US 8,140,271 B2Grant

Least-square deconvolution (LSD): a method to resolve DNA mixtures

Issue Date:2012-03-20

•24 Claims

•46 Drawing Sheets

Abstract

Least Square Deconvolution (LSD) uses quantitative allele peak data, for example, allele peak area, allele peak height and optical density, derived/obtained from a sample containing the DNA of more than one contributor to resolve the best-fit genotype profile of each contributor. The resolution is based on finding the least square fit of the mass ratio coefficients at each locus to come closest to the quantitative allele peak data. Consistent top-ranked mass ratio combinations from each locus can be pooled to form at least one composite DNA profile at a subset of the available loci. The top-ranked DNA profiles can be used to check against the profile of a suspect or be used to search for a matching profile in a DNA database.

Metadata

Assignee

University of Tennessee Research Foundation

Inventors

Tse-Wei Wang
John Douglas Birdwell
Xue Ning
Mark S. Rader
John Flaherty

Application Information

Application Number:US 12/947,063

Filing Date:2010-11-16

Priority Date:2002-10-08

Art Unit:1631

Classifications

IPC:

G06F7/00

Patent Drawings (46 sheets)

Description

[0001] This application is a continuation of U.S. patent application Ser. No. 12/631,032, filed Dec. 4, 2009, now allowed, which is divisional of and claims priority to U.S. patent application Ser. No. 11/413,183, filed Apr. 28, 2006, now U.S. Pat. No. 7,672,789, which is a divisional of and claims priority to U.S. Ser. No. 10/265,908, filed Oct. 8, 2002, now U.S. Pat. No. 7,162,372, which are all incorporated by reference herein in their entirety.

[0002] The U.S. Government retains certain rights in this invention due to funding provided by contract J-FBI-98-083 awarded by the Federal Bureau of Investigation.

FIELD OF THE INVENTION

[0003] The invention is related to methods of resolving a sample containing the DNA of more than one individual into a genotype profile for each individual in the sample.

BACKGROUND OF THE INVENTION

[0004] DNA identification has become one of the most important application tools in forensic science since DNA typing methodologies were introduced around 1985 and may be one of the most important discoveries in the field since the introduction of fingerprinting. With its extremely high capability to differentiate one individual from another (2), it has become widely used in courts around the country and worldwide.

[0005] In recent years, as DNA typing technology has improved and a national DNA database has become available, the popularity and effectiveness of DNA typing methodologies has increased. DNA analysts and other law enforcement agencies have been trying to understand the hidden information contained in DNA through an understanding of the molecular biology, genetics, and statistics involved to provide justice in trials. Because the probability that one person has the same genotype at a set of prescribed DNA loci as another person is very small, DNA typing is widely used in forensic identification, especially when unidentified criminals leave testable evidence such as semen, blood, and saliva at crime scenes. These important stains provide the extracted DNA samples that are to be used for criminal identification.

[0006] DNA typing for forensic applications is based on applying statistical tools to fundamental principles of diagnosis and gene characteristic analysis (2). The DNA profile obtained from criminal evidence has a unique identity, and the characteristics of the DNA profile are analyzed using these methods. The objective of DNA typing is to identify the genotype of the individual who left the evidence. After the perpetrator's genotype is obtained from DNA analysis, forensic caseworkers can compare the genotype of the criminal with that of a suspect or can search for a matching DNA profile in the local, state, and national CODIS databases for a possible suspect (2). Therefore, as an early step in investigation, DNA typing results of forensic samples should be obtained.

[0007] In many cases, especially in rape cases, when a DNA sample is extracted from a biological stain containing body fluids or tissues from more than one person, the result is often a mixed DNA profile. This kind of DNA profile is essentially composed of one contributor's DNA sample superimposed on that of another (3). Much of the DNA evidence obtained from crime scenes is a mixture of more than one contributor's DNA. Generally, the genotype of the victim is known, but the genotype of the perpetrator cannot be obtained clearly and directly due to the presence of DNA of another person in the sample. The genotype of each contributor to the DNA mixture must be deciphered first before further investigation.

[0008] Until now, the deconvolution of mixed DNA profiles contributed by multiple people has been one of the most challenging tasks facing forensic scientists. Part of the difficulty derives from the large number of possible genotype combinations that can be exhibited by the multiple contributors (4) in the mixed DNA profile. So far, no analytical and reliable method has been published for the resolution of DNA mixture into its components.

[0009] Early methods to resolve the genotype profile of contributors in a sample used loci with four alleles to estimate the mass ratio between the two contributors (5). For a locus with four detected alleles, each contributor has to have two different alleles with no shared allele between the two contributors. Therefore, only one allele assignment structure is possible (two heterozygotes). For loci with only two or three alleles more than one possible allele assignment structure is possible at each locus. To determine the genotype profile of an individual at two- or three-allele loci, an initial-guess mass ratio derived from the four-allele loci was used to estimate and evaluate all the possible allele assignment combinations that could be made by the contributors to the sample. The mass ratio at the two- and three-allele loci that best fit the observed relative allele peak areas was identified as the contributor's genotype profiles. This procedure was labor-intensive, and yielded a conservative resolution result.

[0010] More recently, in 1998, the British group of P. Gill et al. of the Forensic Science Services (5) presented a novel method to resolve DNA mixtures using quantitative allele peak data. This method requires an iterative search for the optimum mass ratio to fit the allele peaks at each locus that an individual can contribute to a sample. For each mass ratio used to fit each possible genotype profile, the residuals between the expected allele peak areas and those obtained from the measured allele peaks are calculated. The smallest residual at each locus is added to the minimum residuals similarly derived from allele peak data available at other loci. The genotype combinations that give the overall lowest minimum residual are selected to be the best-fit genotype combinations for the loci. This method is limiting and artificial because a finite set of prior-determined mass ratios is used to calculate the fitting residual. Further, this method is labor intensive because iterations are involved in searching for the best-fit genotype combinations.

[0011] In 2001, Mark Perlin and Beata Szababy developed the Linear Mixture Analysis (LMA) method to resolve DNA mixtures using quantitative allele peak data (18). In this method, all the quantitative allele peak data of all loci in a sample are integrated into a single matrix computation (18). This method imposes the same mass ratio to all loci analyzed in the mixture. This is in contrast to the observation that the best-fit mass ratio may vary from locus to locus in a sample, due to unequal DNA amplification and other nonidealities (24). It is predicted that the imposition of the same weight fractions to fit all loci will present a limitation on that set of weight fractions being optimal for all loci.

[0012] There is a need in the art for an efficient and accurate method to resolve a sample mixture of DNA into the genotype of each individual whose DNA is contained within the mixture.

BRIEF SUMMARY OF THE INVENTION

[0013] The invention encompasses a method of resolving a mixture comprising DNA of more than one individual into genotype profiles for individuals in the mixture. When the method of the present invention is implemented in application software or otherwise, it will be referred to herein as LSD. LSD is an acronym for a mathematical process, in particular, least square deconvolution, which we have picked as the name of the present method, for example, when embodied in software. The use of the acronym LSD is not intended to be limited in describing the present method itself or particular steps of the method for which other steps and known mathematical processes may be substituted by one of skill in the art to equivalent advantage. A step of the method is obtaining quantitative allele peak data at a first locus. A best fit mass ratio coefficient vector is solved using the quantitative allele peak data for allele combinations that can be contributed by the individuals. Residuals are calculated for the allele combinations. An allele combination is selected for the individuals at the first locus having the smallest residual. The smallest residual does not cluster with the second smallest residual. The allele combination selected comprises the genotype profiles of the individuals.

[0014] The invention also encompasses a method of analyzing quantitative allele peak data from a sample comprising DNA of more than one individual into a genotype profile for individuals in the sample. A step of the method is solving for a best fit mass ratio coefficient vector using allele peak data for allele combinations at a first locus that can be contributed by the individuals. Residuals are calculated for the allele combinations. An allele combination for the individuals at the first locus having the smallest residual. The smallest residual does not cluster with the second smallest residual. The allele combination selected comprises the genotype profiles of the individuals.

[0015] The invention further encompasses a method of remotely accessing a software application in a secure manner for resolving a mixture of DNA. The software application is hosted on a secure server. The software application is accessed from a client remotely via a network. The secure server and the client are protected via a firewall. The DNA mixture is transmitted to the secure server. The analysis results are received from the secure server at the client.

[0016] The invention further encompasses a method of generating genotype profiles for individuals who contribute DNA to a sample comprising DNA of more than one individual. A step of the method is obtaining quantitative allele peak data for a set of more than one loci in the sample. The quantitative allele peak data for each locus of the set of loci is separately assigned to allele combination that can comprise the genotype profiles of the individuals at each locus of the set of loci. A residual error and a mass ratio is separately computed for the allele combinations that can comprise the genotype profiles of the individuals at each locus of the set of loci. The allele combinations for each locus of the set of loci are selected. The mass ratio for the allele combinations selected is consistent. The residual error for the allele combinations selected is the smallest or the second smallest residual error and the allele combinations selected comprise the genotype profiles of the individuals who contribute DNA to the sample.

[0017] The invention also encompasses a method of analyzing least square deconvolution output data wherein the data include a mass ratio and residual for allele combinations at a first locus in a set of loci in a sample comprising DNA of two individuals. A step of the method is preliminarily selecting either a genotype combination for the two individuals having a residual that is smallest if the smallest residual does not cluster with the second smallest residual or preliminarily selecting more than one genotype combination for the two individuals if the more than one genotype combination comprises residuals that are the smallest and that cluster. The genotype combination for the two individuals from the preliminarily selected combination are determined where the genotype combination has a mass ratio consistent with that of a second locus determined for the sample.

BRIEF DESCRIPTION OF THE DRAWINGS

[0018] FIG. 1 is a flow diagram of the LSD method to resolve mixture DNA samples. For each locus, the best fitting genotypes for the two contributors and the approximate mass ratio can be obtained in one step.

[0019] FIG. 2 is a flow diagram to show the typical processing steps of using the ABI 310 Genetic Analyzer to identify alleles present at each locus.

[0020] FIG. 3 shows the typical output of a DNA profile using the ABI 310 Genetic Analyzer. Each peak corresponds to one allele, and the peak area is proportional to the mass of the allele it represents.

[0021] FIG. 4 shows a preferred embodiment of the invention in which LSD is implemented using software running under a secure web server.

[0022] FIG. 5 is an illustration of allele peaks of a typical heterozygous locus from a single source.

[0023] FIG. 6 shows simulated superimposed allele peaks at one locus for a mixed DNA profile, with the contributors' genotypes of [3, 15] and [7,15], mixed at a mass ratio of 1:0.4 respectively. Note that allele 15 is the shared allele.

[0024] FIG. 7 shows the projection of b vector (4,6) onto A vector (9,5). The error vector (b−{circumflex over (b)}) is perpendicular to the A vector and gives the shortest distance from b to A.

[0025] FIG. 8 shows vector b is in the column space of A (3×2 matrix). Therefore, b can be expressed as a linear combination of columns of A.

[0026] FIG. 9 shows the projection of b vector onto the column space of a 3 by 2 matrix A. p is the projection of b vector onto the column space of A.

[0027] FIG. 10 is a flow diagram with an overview of the major steps in putting together a composite genotype using LSD.

[0028] FIG. 11 is a flow sheet of using LSD to resolve mixed DNA at a 2-allele locus.

[0029] FIG. 12 shows the allele peaks at twelve loci of Texas DNA mixture data number 1; both the allele name and the corresponding peak areas are indicated.

[0030] FIG. 13A, B, and C show the output of web-LSD for Texas DNA mixture data number 1. All possible genotype combinations are ranked at each locus according to the magnitudes of the fitting errors.

[0031] FIG. 14 shows the normalized fitting residuals of all possible genotype combinations at the locus VWA in the DNA mixture data.

[0032] FIG. 15 shows the normalized fitting residuals of all possible genotype combinations at the locus D21S11 in the DNA mixture data.

[0033] FIGS. 16A and B show the output of web LSD for the simulated DNA mixture data with a mass ratio of 1:5. All possible genotype combinations are ranked at each locus according to the magnitudes of the fitting errors.

[0034] FIGS. 17A and B show the output of web LSD for the simulated DNA mixture data with a mass ratio of 1:2. All possible genotype combinations are ranked at each locus according to the magnitudes of the fitting errors.

[0035] FIGS. 18A and B show the output of web LSD for the simulated DNA mixture data with a mass ratio of 1:1. All possible genotype combinations are ranked at each locus according to the magnitudes of the fitting errors.

[0036] FIGS. 19A and B show the output of web LSD for the DNA mixture data presented in Clayton et al. (3). All possible genotype combinations are ranked at each locus according to the magnitudes of the fitting errors.

[0037] FIGS. 20A and B show the output of web LSD for the DNA mixture data presented in Evett et al. (21). All possible genotype combinations are ranked at each locus according to the magnitudes of the fitting errors.

[0038] FIGS. 21A, B, C, and D show the output of web LSD for Acadiana DNA mixture number 5. All possible genotype combinations are ranked at each locus according to the magnitudes of the fitting errors.

[0039] FIGS. 22A, B, C, and D show the output of web LSD for Acadiana DNA mixture number 1. All possible genotype combinations are ranked at each locus according to the magnitudes of the fitting errors.

[0040] FIGS. 23A, B, and C show the output of web LSD for Acadiana DNA mixture number 2. All possible genotype combinations are ranked at each locus according to the magnitudes of the fitting errors.

[0041] FIG. 24 shows the allele peaks at nine loci of the Texas DNA mixture data number 3; both the allele names and the corresponding peak areas are indicated.

[0042] FIGS. 25A, B, and C show the output of web LSD for Texas DNA mixture number 3. All possible genotype combinations are ranked at each locus according to the magnitudes of the fitting errors.

[0043] FIG. 26 shows the allele peaks at nine loci of the Texas DNA mixture data number 4; both the allele names and the corresponding peak area are indicated.

[0044] FIGS. 27A, B, and C show the output of web LSD for Texas DNA mixture number 4. All possible genotype combinations are ranked at each locus according to the magnitudes of the fitting errors.

DETAILED DESCRIPTION OF THE INVENTION

[0045] To date, no direct, systematic, analytic, and quantitative method exists to resolve DNA mixture samples. The instant invention, based on quantitative allele peak data, provides the art with a method to resolve DNA mixture samples contributed by two individuals.

[0046] Least Square Deconvolution (LSD) is a novel method applying the least-square modeling approach to find the best-fit genotype combination to resolve mixed DNA samples comprising DNA of more than one person. Quantitative allele peak data are used in LSD because allele peak areas are theoretically proportional to the mass of the corresponding DNA alleles in a mixture and because the proportional relationship of allele peak areas is approximately preserved during PCR amplification. Other types of measurements with a known theoretical relationship to DNA allele mass in a sample and that approximately preserve this relationship in practice are equivalent. Examples of such equivalent types of measurement include allele peak height and optical density. The theoretical relationship does not have to be linear as long as it is known. The objective of LSD is to first find the best-fit genotype combination for the two contributors at loci where peak data are available, using least-square techniques and the measured allele peak data. Then, using the best fit mass ratio information for all loci processed, a composite genotype profile for each of the individual contributors can be formed that is compatible with the results of the least square analysis.

[0047] The advantage of LSD compared to other quantitative approaches is its direct calculation of the best-fit genotype combination and the approximate mass ratio, without iterative searching for the optimal mass ratio. See (5) and (18). Some matrix calculations are involved in this method. FIG. 1 is a flow diagram of an embodiment of LSD. It is apparent from the diagram that LSD is applied at each locus independently of other loci, thus allowing each locus to be fitted with its own best-fit mass ratio, which can differ from the best-fit mass ratio arrived at other loci.

[0048] Compared with other approaches published in the literature, LSD is more efficient, simpler, more comprehensive, and gives true genotypes when the quantitative allele peak data and its theoretical relationship to allele mass approximately preserve the relative DNA mass proportionality. The advantage of LSD is its direct calculation of the best-fit genotype combination and the approximate mass ratio without iteration (5). Furthermore, LSD is applied to each locus independently of other loci, thus allowing each locus to be fitted independently with its own best-fit mass ratio, and the degree of confidence of fit to be independently assessed for each locus. From examining the relative errors of each fit at a locus, the degree of confidence can be separately assigned to the resulting best-fit genotype, locus by locus, allowing a composite profile to be assembled with a high degree of confidence containing only those loci whose LSD results are clear cut.

[0049] A locus refers to the position occupied by a segment of a specific sequence of base pairs along a gene sequence of DNA (2). Genes are differentiated by their specific sequences of base pairs at each locus. An allele refers to the specific gene sequence at a locus. At most two possible alleles can be present at one locus of a chromosome pair for each individual: one contributed by the paternal and the other contributed by the maternal source (8). If these two alleles are the same, the DNA profile is homozygous at that locus. If these two copies are different, the DNA profile is heterozygous at the locus (8). There are multiple alleles that can be contributed by either parent at each locus.

[0050] A genotype or DNA profile is the set of alleles that an individual has at a given locus. A genotype or DNA profile may also comprise the sets of alleles that an individual has at more than one locus. For example, a genotype or DNA profile may comprise the set of alleles at each of at least 2 loci, 3 loci, 4 loci, 5 loci, 7 loci, 9 loci, 11 loci, 13 loci, or 20 loci.

[0051] A DNA or genotype profile is developed from a nucleic acid sample, usually a DNA sample. Sources of nucleic acid include tissue, blood, semen, vaginal smears, sputum, nail scrapings, or saliva.

[0052] The DNA of interest can be prepared for analysis by the LSD method by amplification and subsequent separation. Amplification may be performed by any suitable procedures and by using any suitable apparatus available in the art. For example, enzymes can be used to perform an amplification reaction, such as Taq, Pfu, Klenow, Vent, Tth, or Deep Vent. Amplification may be performed under modified conditions that include “hot-start” conditions to prevent nonspecific priming. “Hot-start” amplification may be performed with a polymerase that has an antibody or other peptide tightly bound to it. The polymerase does not become available for amplification until a sufficiently high temperature is reached in the reaction. “Hot start” amplification may also be performed using a physical barrier that separates the primers from the DNA template in the amplification reaction until a temperature sufficiently high to break down the barrier has been reached. Barriers include wax, which does not melt until the temperature of the reaction exceeds the temperature at which the primers will not anneal nonspecifically to DNA.

[0053] The products of the amplification reaction are detected as different alleles present at a locus or loci. The alleles of at least one locus are amplified and detected after the amplification reaction. If desired, however, the alleles of multiple loci, e.g., two, three, four, five, six, ten, fifteen, twenty, twenty-five, or thirty, or more different loci may be detected after amplification. Sets of loci may include at least two, three, five, ten, fifteen, twenty, thirty, or fifty loci. Amplification of all of the alleles may be performed in a single amplification reaction or in a multiplex amplification reaction. Alternatively, the sample may be divided into several portions, each of which is amplified with primers that yield product for the alleles present at a single locus. Multiplex amplification is preferred.

[0054] The different alleles at a locus typically are detected because they differ in size. Alleles can differ in size due to the presence of repeated DNA units within loci. A repeated unit of DNA can be, e.g., a dinucleotide, trinucleotide, tetranucleotide, or pentanucleotide repeat. Short Tandem Repeats (STR) are DNA segments with repeat units of 2-6 bp in length (10). The repeated unit can be of a longer length that ranges from ten to one hundred base pairs. These are medium-length repeats and may be referred to as a Variant Number of Tandem Repeat (VNTR) (10). Repeat units of several hundred to several thousand base pairs may also be present in a locus. These are the long repeat units.

[0055] The number of repeated units at a locus also varies. The number of repeated units may be, for example, at least five, at least ten, at least fifteen, at least twenty, at least twenty-five, or at least fifty units. The effect of these repeated units of DNA is the presence of multiple types of alleles that an individual can possess at any given locus that can be detected by size (10).

[0056] Preferably, alleles that harbor different numbers of STR repeat units are detected. More than 8000 STRs (loci) scattered across the 23 pairs of human chromosomes have been collected in the Marshfield Medical Research Foundation in Marshfield, Wis. (10). Preferably, alleles at the 13 core loci used by the FBI Combined DNA Index System (CODIS): CSF1PO, FGA, TH01, TPOX, VWA, D3S1358, D5S818, D7S820, D8S1179, D13S317, D16S539, D18S51, and D21S11 (11), are detected.

[0057] It is also contemplated that amplification may be performed to detect an allele by amplifying microsatellite DNA repeats, DNA flanking Alu repeat sequences, or any other known polymorphic region of DNA that can be distinguished based on the size of different alleles.

[0058] Any method that separates amplification products based on size and any method that quantitates the amount of the allele present in the sample can be used to prepare the data required for analysis of genotype profiles in the method. The amplification products may be separated by electrophoresis in a gel or capillary, or mass spectrometry. The amount of each allele present may be determined flourometrically in a flourometer, or via ultraviolet spectrometry. For example, a Beckman Biomek® 2000 Liquid Handling System can be used to detect and quantitate alleles present for a locus in a sample. Optical density or optical signal can be used to detect the presence of an allele after gel or capillary electrophoresis.

[0059] Preferably, alleles are detected using an ABI Prism 310 Genetic Analyzer, or a HITACHI FMBIO II Fluorescence Imaging System (10). The ABI 310 Genetic Analyzer identifies alleles present at a locus, as outlined in FIG. 2, and provides a data output result, as shown in FIG. 3. One advantage of this instrument is that, in addition to sizing the detected allele signals, the related software can also display their peak heights and automatically calculate the area under each peak (10).

[0060] The HITACHI FMBIO II Fluorescence Imaging System uses gel electrophoresis instead of capillary electrophoresis to separate the alleles of a DNA sample (10). This system requires much more sample and a longer time to complete a separation. In this genetic analyzer, each allele corresponds to a specific band in a gel lane. The band size for each allele is compared with a well-calibrated allelic ladder to identify the corresponding allele (10).

[0061] If the amplification products are input into an apparatus that both separates and quantitates alleles for a locus in a sample, four different types of peaks can be obtained from these raw data: true or allele peaks, stutter peaks, artifact peaks, and pull up peaks.

[0062] True or allele peaks are peaks that indicate the presence of an allele at a locus. The most important characteristic of an allele peak is that the measured peak area or height is roughly proportional to the mass of the corresponding allele in the DNA sample (10). Preferably peak area is used.

[0063] Stutter peaks are peaks generated by the enzyme's slippage during the amplification process (12). In most cases, stutter peaks are located on the left side of the associated alleles, and the gene distance between the stutter peak and the associated allele peak is usually less than 4 bp (12). The height of the stutter peak is usually less than 15% of the height of the corresponding true allele peak (12).

[0064] Artifact peaks are peaks due to impurities in the DNA samples. Generally, the artifact peaks have one or more of the following three characteristics: (1) about 53% of them are less than 5% of the nearest allele peak's height (12), (2) some artifact peaks consist of multiple peaks, and the distances among them are always less than 1 bp (12), and (3) some artifact peaks are within 0.5 bp of an allelic ladder marker (12). If a peak satisfies any of the above three rules, the peak can be defined as an artifact peak, and the peak's effect can be eliminated.

[0065] A pull-up peak is a minor peak directly to the right of a ‘true’ allele peak. Usually, a pull-up peak is located on the right side of a ‘true’ allele peak with a distance less than ⅜ bp, and its height is less than 50% of the major peak.

[0066] Quantitative peak data of ‘true’ alleles are determined at a locus. These measurements may be the peak height or peak area of a signal detected by an instrument or procedure designed to quantify the presence of each allele. The peak height, peak area, and any other measurement that is related to the relative masses of each allele present in the original stain or sample are equivalent. Quantitative allele peak data will be referred to as “peak height,” “peak area,” or “quantitative allele peak data.” Each of these terms is interchangeable.

[0067] The allele peaks or areas are calculated and analyzed using LSD. LSD then returns the “best-fit” genotype profiles for the two individuals that contribute to the sample. “Best-fit” refers to an assumption that the allele peak area/height is proportional to the relative mass proportion of the corresponding DNA allele in the mixture, the returned genotypes at the specified mass proportions would yield a set of allele peak areas/heights that is ‘closest’ to the measured set of allele areas/heights, in the least square sense (as measured by the Euclidean distance metric).

[0068] The genotype profile assigned to each individual by the LSD method can be verified by comparing the known genotype profile of one individual that contributed to the sample to that of one person developed by the LSD method. The known genotype profile may be obtained from an individual that is the victim of a crime.

[0069] A genotype profile obtained from the LSD method may also be matched to an individual to identify the individual as potentially having contributed to the sample. The genotype profile may be matched to the individual after obtaining a sample from the individual. The genotype profile may also be matched to an individual by comparing it to other genotype profiles in a database. The database may be any public or proprietary database that stores and/or matches genotype profiles. The database may be CODIS, which may be used to store genotype profiles in a national, state, or regional collection, and which may separate these profiles into disjoint parts, such as a convicted offenders database, a forensic DNA database, or a missing persons database.

[0070] A preferred embodiment of the invention is shown in FIG. 4. In this embodiment, LSD is implemented using software running under a secure web server 1 on a protected network 2 that is isolated from a public or private network 3 by a firewall 4. A remote user located at LSD/Database Client station 8 may access the LSD software at the web server 1 via the public or private network. The communication may be via the public switched telephone network (PSTN) preferably using known encryption algorithms for confidential data but is preferably via a private network and encrypted. The firewall 4 allows communications with the secure web server 1 using an encrypted communications protocol such as the Hypertext Transfer Protocol (HTTP) over a Secure Sockets Layer (SSL). The firewall 4 connects the protected network 2 to the public or private network 3 using either an Internet service provider (ISP), leased, or owned telecommunications equipment/circuits 5 having appropriate bandwidth capability (although the data may be suitably compressed via known compression algorithms and transmitted over lower bandwidth facilities). The connection to the firewall 4 and all connections and equipment collocated with the protected network 2 are housed in a secure server facility 6 that provides LSD services to a community of clients located at forensic laboratories 7 or other organizations. Location 7, 8, 9 is shown by way of example only and is no way intended to be limited to forensic laboratory locations.

[0071] A client 8 located at a forensic laboratory or other organization may use the public or private network 3 to gain access to LSD services offered by the secure server facility 6. Preferably, the client 8 is connected to a protected network 9 which connects to the public or private network 3 through a firewall 10, and the firewall 10, the protected network 9, and all equipment connected to the protected network 9, such as the LSD/Database Client 8, are housed in a secure client facility such as a forensic laboratory 7 (or other secure facility). The firewall 10 located at the forensic laboratory 7 connects the protected network 9 to the public or private network 3 using either an ISP, leased, or owned telecommunications equipment/circuits 11 having similar bandwidth considerations as described above for equipment/circuits 5.

[0072] The client 8 may make requests to analyze data derived from DNA mixtures on the secure LSD web server 1 by accessing the secure web server 1, transmitting DNA mixture data to the secure web server, and receiving analysis results. These results may then be interpreted using mixture interpretation guidelines to obtain one or more DNA profiles that may be associated with a suspect to a crime.

[0073] Optionally, the LSD/Database Client 8 may access a local laboratory, state, or national DNA database 12 to search for matches to the one or more DNA profiles formed using the results of the LSD analysis. The DNA database 12 may be located in a separate secure facility at the state, local, or national level and is preferentially protected by a firewall 13. The firewall 13 is connected to the public or private network using either an ISP, leased, or owned telecommunications equipment/circuits 14, and preferentially allows communications with a DNA database server 12 using only an encrypted communications protocol such as HTTP over SSL. The firewall 13 and DNA database server 12 are connected to a protected network 15. The connections to the firewall 13 and all connections and equipment collocated with the protected network 15 are housed in a secure server facility 16 that provides DNA database services to a community of clients located at forensic laboratories 7 or other organizations.

[0074] Nothing shown in FIG. 4 or described above should be taken to restrict the domain of the invention. For example, the DNA database server and the secure LSD server may be connected through firewalls to two separate and isolated public or private networks, requiring a separate client and protected network located at a forensic laboratory in order to communicate with each server. This is the case at present with the FBI's National DNA Index System (NDIS), which is connected to state and local facilities through the FBI-owned and operated Criminal Justice Information System's Wide Area Network (CJIS-WAN), and with the current implementation of the secure LSD server. This server is located on a protected network within The University of Tennessee's Laboratory for Information Technologies (LIT) and is connected through a firewall owned and operated by LIT to the university's campus network and thence to the public Internet. In this case, the functionality remains the same, except that an investigator or analyst transfers results obtained by a client from the secure LSD server to a client computer of the FBI's NDIS facilities in order to perform a search on the national DNA database.

[0075] The invention is not restricted to operation on protected computers and networks, nor is it restricted to require security of communications using encryption and secure authentication protocols. However, these measures are usually necessitated by the privacy laws of the United States and other countries. In a similar manner, it is not required that the LSD software, LSD/Database Client, and DNA database software operate on separate and communicating computers. They may in fact all be installed and operated on a single computer in some applications, or on two computers. There may also be multiple instances of the DNA database software running on several computers. The realities of multiple jurisdictions and multiple ownership of and responsibility for controlled access to data that are considered sensitive usually necessitates the use of multiple computers under the control of independent but cooperating agencies.

EXAMPLES

Example 1

Identification of DNA Mixtures with More than One Contributor and Determination of the Number of Contributors in the DNA Mixtures

[0076] According to studies reported in the literature (10), if only one person contributes to a DNA sample, then no more than two alleles will be detected at a locus. (There have been cases where a DNA sample from one person does cause detection of three alleles at a locus, but these cases are extremely rare.) Therefore, a DNA/STR sample is highly likely to be a mixture if more than two alleles are consistently present at multiple loci. For a locus with two heterozygous alleles from a single contributor, if the higher peak is set as 100%, the heterozygous companion allele, due to the co-amplification of the two alleles, usually has a peak area or peak height greater than 70% of the higher allele (10). A severe peak-height imbalance between any two heterozygous alleles at a locus strongly suggests that the DNA sample being tested is a mixture. FIG. 5 is an illustration of a typical heterozygote DNA/STR profile from a single contributor. Note that only two ‘true’ alleles are present; the two minor peaks are considered to be stutter peaks (10), an artifact of the DNA amplification procedure.

[0077] If a sample comes from only one person, usually no more than two true alleles can be detected at one locus. If there are still more than two peaks appearing at several loci, even after eliminating the effects of the stutter peaks, the pull-up peaks, and the artifact peaks, it can be concluded that the sample contains a mixture from at least two separate contributors.

[0078] Because usually no more than two alleles can be detected at one locus for a one-person DNA sample, the maximum number of alleles that typically can be detected at one locus for a two-person mixed DNA/STR profile is four. Five or more alleles present at several loci would be strongly indicative of three or more contributors (3). In the national CODIS DNA database in 2002, approximately 9% of DNA profiles are mixture samples, but the DNA analysts spend 90% of their efforts and time trying to interpret them (13). In most forensic cases, especially in rape cases, a great number of mixed DNA samples are contributed by two individuals, although three-people mixtures are also encountered occasionally (3).

[0079] Some alleles may be shared by each of the contributors to a DNA mixture. Thus, the peaks in the mixture sample can be regarded as the superposition, or sum of the two individuals' allele peaks. These important characteristics allow one to resolve the DNA mixture profile by using the allele peak information. FIG. 6 shows an artificial example of allele peaks contributed by two people. Note that only two heterozygous alleles associated with the same contributor have equal peak areas (in the ideal case), and allele 15 is shared between the two individuals. In addition, the peak area of alleles from person 2 is about 40% of that from person 1.

Example 2

Mathematical Principles for Solving the Linear Least-Square Solution to Data Fitting

[0080] To perform LSD, multivariable linear equations (Ax=b) are solved for data fitting. Most overdetermined multivariable linear equations do not have exact solutions because they are inconsistent with each other. In those cases, the least-square approach can be used to obtain a best-fit solution (16), which yields a fitted {circumflex over (b)} that is closest to the given b.

Example 2A

Introduction to the Least-Square Approach

[0081] Let A be an in by n (m>n) matrix and b be a column vector composed of in elements. For a linear equation of Ax=b, an exact solution for this equation, x, can be obtained only when b is in the column space of A. (The vector b can be expressed as a linear combination of the columns of the matrix, A. The elements in the vector, x, determine the linear combination.) The multiplication of A and x can be viewed as:

[0082]

\begin{matrix} Ax = b \Rightarrow [\begin{matrix} | & | & | & | \\ | & | & | & | \\ a_{1} & a_{2} & \dots & a_{n} \\ | & | & | & | \\ | & | & | & | \end{matrix}] [\begin{matrix} x_{L} \\ ⋮ \\ ⋮ \\ x_{n} \end{matrix}] \\ = [\begin{matrix} b_{1} \\ ⋮ \\ ⋮ \\ ⋮ \\ b_{m} \end{matrix}] \Rightarrow x_{1} [\begin{matrix} | \\ | \\ a_{1} \\ | \\ | \end{matrix}] + x_{2} [\begin{matrix} | \\ | \\ a_{2} \\ | \\ | \end{matrix}] + \dots + x_{n} [\begin{matrix} | \\ | \\ a_{n} \\ | \\ | \end{matrix}] \\ = [\begin{matrix} b_{1} \\ ⋮ \\ ⋮ \\ ⋮ \\ b_{m} \end{matrix}] . \end{matrix}

[0083] For example: if A is

[0084]

[\begin{matrix} 1 & 2 \\ 6 & 5 \\ 3 & 4 \end{matrix}],

and b is

[0085]

[\begin{matrix} 8 \\ 27 \\ 18 \end{matrix}],

the linear equation:

[0086]

Ax = b ([\begin{matrix} 1 & 2 \\ 6 & 5 \\ 3 & 4 \end{matrix}] x = ([\begin{matrix} 8 \\ 27 \\ 18 \end{matrix}])

has an exact solution x, because b can be expressed as a linear combination of the columns of A as shown:

[0087]

[\begin{matrix} 1 \\ 6 \\ 3 \end{matrix}] 2 + [\begin{matrix} 2 \\ 5 \\ 4 \end{matrix}] 3 = [\begin{matrix} 2 + 6 \\ 12 + 15 \\ 6 + 12 \end{matrix}] = [\begin{matrix} 8 \\ 27 \\ 18 \end{matrix}] \Rightarrow {\begin{matrix} x_{1} = 2 \\ x_{2} = 3. \end{matrix}

Therefore, the solution for this multivariable linear equation is

[0088]

x = [\begin{matrix} 2 \\ 3 \end{matrix}],

where A and b are given as stated.

[0089] When b cannot be expressed as a linear combination of the columns of A, as shown in the example described below, then no solution, x, would exist that exactly satisfies the equation Ax=b. Such system is termed inconsistent and has no solution (16). For example, the simultaneous equation:

[0090]

{\begin{matrix} 9 x = b_{1} \\ 5 x = b_{2}, \end{matrix}

which can also be written as:

[0091]

Ax = b (A = [\begin{matrix} 9 \\ 5 \end{matrix}]),

can be solved with exact solution only when b₁and b₂are in the ratio of 9:5, which means that the exact solution x exists only when b is on the same line as the vector A

[0092]

(such as [\begin{matrix} 9 \\ 5 \end{matrix}])

drawn from the origin. If b is

[0093]

[\begin{matrix} 4 \\ 6 \end{matrix}],

it is clear that no exact solution x can fit both of these two equations simultaneously. Seemingly unsolvable equations such as this are often encountered in practice from modeling experimental results, and must still be solved. In this case, an ‘approximate’ solution is to be sought, such that A times the approximate solution {circumflex over (x)}, would give a {circumflex over (b)} that is ‘closest’ to the given b. For the example given above, from FIG. 7, one can see that the distance between the given b vector and {circumflex over (b)}, the projection of b vector onto the A vector is the shortest among all the routes from b to the line A. This implies that the error vector (b−{circumflex over (b)}) for the optimal b must be perpendicular to the A vector. Since A{circumflex over (x)}={circumflex over (b)}, we require the inner product between the A vector and the error vector to be 0: A^T(b−{circumflex over (b)})=A^T(b−A{circumflex over (x)})=0 custom character

A^TA{circumflex over (x)}=A^Tb. The transpose of the matrix A is denoted by A^T. The solution is given by {circumflex over (x)}=(A^TA)⁻¹A^Tb if (A^TA)⁻¹exists. If not, the pseudoinverse of A must be used. In a single variable case, {circumflex over (x)} is given by {circumflex over (x)}=A^Tb/(A^TA) because A^TA is a scalar. For the example in FIG. 7, the {circumflex over (x)} is given by:

[0094]

\hat{x} = A^{T} b / (A^{T} A) = \frac{9 * b_{1} + 5 * b_{2}}{9 * 9 + 5 * 5} .

The corresponding {circumflex over (b)} is given by:

[0095]

\hat{b} = [\begin{matrix} 9 \\ 5 \end{matrix}] \hat{x} .

[0096] Another approach to arrive at the optimum {circumflex over (x)} minimizes the sum of the squares of the fitting error between the calculated {circumflex over (b)} and the measured data b using calculus. For the example above, the error can be expressed as:
E²=(9{circumflex over (x)}−b₁)²+(5{circumflex over (x)}−b₂)².

[0097] The minimum error E²would be zero, if an x can be found to fit both of these two equations exactly. In most cases, the vectors b and a are not in the same line, and the value of E²would not be equal to zero. E²is expressed as a function of {circumflex over (x)}, and the minimum value of E²can be obtained by taking the derivative of E²with respect to {circumflex over (x)} i.e. d(E²)/d{circumflex over (x)}, and setting it to 0, followed by solving for {circumflex over (x)}. {circumflex over (x)} is found to be:

[0098]

\hat{x} = \frac{9 * b_{1} + 5 * b_{2}}{9 * 9 + 5 * 5} .

[0099] The {circumflex over (x)} that minimizes E²is referred to as the least-square solution to the inconsistent system Ax=b. The solution {circumflex over (x)}, arrived by the two approaches, is exactly the same.

Example 2B

Multivariable Least-Square Solution to Inconsistent Linear System of Equations, with More Equations than Unknowns

[0100] The multivariable system of equations, Ax=b, which has n variables and m observations, is solvable only when the vector b is a linear combination of the columns of A. FIG. 8 shows an example where the vector b is in the column space of A (a 3 by 2 matrix). In this case, the multivariable equations Ax=b is solvable. When n is less than in, it is very possible that Ax=b is inconsistent, which means that the given vector b does not lie in the space spanned by the columns of A, as shown in FIG. 9. In this case, an exact solution x, to the multivariable systems of equations Ax=b, does not exist. However, an approximate solution can be obtained by choosing {circumflex over (x)}, such that A{circumflex over (x)} comes ‘closest’ to the given vector b. In other words, {circumflex over (x)} would minimize the square of the distance between the given b vector (or, equivalently, the distance), and the fitted point of (A{circumflex over (x)}), which is expressed as
E²=∥A{circumflex over (x)}−b∥².

[0101] Here, ∥·∥ denotes the norm of a vector (or the magnitude of a vector), and can be defined as the square root of the sum of squares of the elements in a vector. Other equivalent definitions of the norm of a vector are possible and can be used. Also, the same solution is obtained if the square is omitted. This least-square approach is widely accepted as an applicable way for multivariable data fitting (16). From FIG. 9, one can see that the perpendicular distance between the vectors b and p, which is the projection of b onto the column space of A, is the minimum of all distances from b to the plane spanned by the two columns of A. In this case, the result of A{circumflex over (x)} gives the projection of the vector b onto the column space of A, or the point p, which is indeed closest to b of all points on the plane. The vector (b−A{circumflex over (x)}) is the error vector e and is perpendicular to the column space of A, This implies:
A^T(b−A{circumflex over (x)})=0.
So, the least-square solution, {circumflex over (x)}, of Ax=b is one that satisfies the so-called ‘normal equation’:
A^TA{circumflex over (x)}=A^Tb.
and the least-square solution for such equation is given by:
{circumflex over (x)}=(A^TA)⁻¹A^Tb.

[0102] If the columns of A are linearly dependent and (A^TA) is not invertible, then the best-fit {circumflex over (x)} is to be computed by the pseudoinverse of x, or x⁺:
{circumflex over (x)}=A⁺b,
where A⁺ represents the pseudoinverse of the matrix A, which is preferably computed using the singular-value-decomposition (SVD) of A (16):
A=UΣV^T,
where U (m by m matrix) and V (n by n matrix) are orthogonal matrices, and Σ (m by n matrix) is a diagonal matrix with non-negative diagonal elements arranged in a decreasing order. The diagonal elements of Σ, σ₁, σ₂, . . . σ_n, are called the singular values of A, and the rank of A is determined by the number of nonzero singular values. The pseudoinverse of A is computed as:
A⁺=VΣ⁺U^T,
where Σ⁺ is an n by m matrix with 1/σ₁, 1/σ₂, . . . 1/σ_ron its main diagonal, where r denotes the number of nonzero singular values; the remaining diagonal elements of Σ⁺, corresponding to those zero singular values in Σ, are also zero. This is one of the hallmarks of the pseudoinverse which enables the least-square solution of an inconsistent system of linear equation to be obtained in one direct step.

[0103] If the columns of A are linearly independent, we can also calculate the least-square solution as:
{circumflex over (x)}(A^TA)⁻¹A^Tb=(VΣ^TΣV^T)⁻¹VΣ^TU^Tb=V(Σ^TΣ)⁻¹V⁻¹VΣ^TU^Tb=VΣ⁺U^Tb=A⁺b.

[0104] If A is a square and full-rank matrix, implying A is invertible, then the equation Ax=b is consistent with the exact solution, x, which is given by:
x=A⁻¹b.

[0105] In this case, U, V, and Σ are all square matrices, and Σ is invertible, and the inverse of A can be calculated as:
A⁻¹(UΣV^T)⁻¹=VΣ¹U^T,
which is exactly the same as A⁺. Therefore, for all multivariable systems of equations Ax=b, the least-square solution, {circumflex over (x)}, can always be calculated from {circumflex over (x)}=A⁺b using the SVD.

Example 3

Principles for Using the Least Square Approach (LSD) to Resolve Mixed DNA Samples into Each Contributor's Genotype

Example 3A

Represent all Possible Genotype Combinations at Each Locus with Matrices

[0106] The allele peak data of a DNA mixture sample, when available, usually includes the allele peak information from 6 to 18 loci. One locus is to be analyzed at a time. For each locus, the number of alleles can be determined by observing all the true peaks in the electropherogram at this locus (3). According to the number of true alleles detected, all possible allele assignment patterns for the two contributors are to be first enumerated. For each possible allele combination pattern for the two contributors, a matrix of coefficients with two columns, one for each person, can be formed. This matrix is referred to as the ‘A’ matrix, such as those shown in Table 1. Each column of A represents the genotype of one contributor at this locus. Each element in the matrix A is either 0, 1, or 2, representing the absence or presence of the corresponding allele. A ‘2’ stands for the homozygote allele indicating that the contributor has two copies of this allele at the specified locus. The number of rows of the matrix A corresponds to the number of alleles observed at this locus in the mixture sample.

[0107]

TABLE 1

	Genotype Combination		Pseudoinverse of the
Case	Person 1	Person 2	Matrix (A)	Matrix (A⁺)

1	A, A	B, C	[ 2 0 0 1 0 1 ]	1 2 ⁡ [ 1 0 0 0 1 1 ]

2	B, B	A, C	[ 0 1 2 0 0 1 ]	1 2 ⁡ [ 0 1 0 1 0 1 ]

3	C, C	A, B	[ 0 1 0 0 2 1 ]	1 2 ⁡ [ 0 0 1 1 1 0 ]

4	A, B	B, C	[ 1 0 1 1 0 1 ]	1 3 ⁡ [ 2 1 - 1 - 1 1 2 ]

5	A, B	A, C	[ 1 1 1 0 0 1 ]	1 3 ⁡ [ 1 2 - 1 1 - 1 2 ]

6	A, C	B, C	[ 1 0 0 1 1 1 ] ⁢	1 3 ⁡ [ 2 - 1 1 - 1 2 1 ]

[0108] All possible genotype combinations and their corresponding matrices at a three-allele locus for the two contributors. The three alleles are designated as {A, B, C}. The numbers {0, 1, 2} in the matrix A represents 0, 1, or 2 copies of the corresponding allele it represents.

[0109] For a DNA mixture contributed by two people, the number of alleles that can be present at one locus is between one and four. If only one allele is present at a certain locus, then undoubtedly both contributors are homozygous with this allele at this locus. In Tables 1, 2, and 3, all possible genotype combinations for the two source contributors and their corresponding ‘A’ matrices are listed for the cases of 3, 2, and 4 alleles present at a locus, respectively. The last column in each of these three tables lists the pseudoinverse matrix of the corresponding genotype matrix ‘A’. The pseudoinverse matrix will be used to compute directly the best-fit mass coefficients during the least-square fitting step.

[0110]

TABLE 2

	Genotype Combination		Pseudoinverse of
Case	Person 1	Person 2	Matrix (A)	the Matrix (A⁺)

1	A, A	B, B	[ 2 0 0 2 ]	1 2 ⁡ [ 1 0 0 1 ]

2	A, B	A, A	[ 1 2 1 0 ]	1 2 ⁡ [ 0 2 1 - 1 ]

3	A, B	B, B	[ 1 0 1 2 ]	1 2 ⁡ [ 2 0 - 1 1 ]

4	A, B	A, B	[ 1 1 1 1 ]	1 4 ⁡ [ 1 1 1 1 ]

[0111] All possible genotype combinations and their corresponding matrices at a two-allele locus for the two contributors. The two alleles are designated as {A, B}. The numbers {0, 1, 2} in the matrix A represents 0, 1, or 2 copies of the corresponding allele it represents.

[0112]

TABLE 3

	Genotype Combination		Pseudoinverse of
Case	Person 1	Person 2	Matrix (A)	the Matrix (A⁺)

1	A, B	C, D	[ 1 0 1 0 0 1 0 1 ]	1 2 ⁡ [ 1 1 0 0 0 0 1 1 ]

2	A, C	B, D	[ 1 0 0 1 1 0 0 1 ]	1 2 ⁡ [ 1 0 1 0 0 1 0 1 ]

3	A, D	B, C	[ 1 0 0 1 0 1 1 0 ]	1 2 ⁡ [ 1 0 0 1 0 1 1 0 ]

[0113] All possible genotype combinations and their corresponding matrices at a four-allele locus for the two contributors. The four alleles are designated as {A, B, C, D}. The numbers {0, 1} in the matrix A represent 0 or 1 copy of the corresponding allele it represents.

[0114] A normalized vector ‘b’ is to be generated from the measured allele peak area data, such that the smallest allele peak area in the vector is normalized to 1, and those of all other alleles in the same locus are ratioed to it. The number of elements in the vector ‘b’ reflects the number of alleles present at this locus. For example, if the peak area measured at one locus is [3454, 4588, 8766], then the corresponding normalized b vector would be:

[0115]

b = [\begin{matrix} 3454 / 3454 \\ 4588 / 3454 \\ 8766 / 3454 \end{matrix}] = [\begin{matrix} 1 \\ 1.33 \\ 2.54 \end{matrix}] .

Other normalizations such that the smallest peak ratios are not equal to one are possible and equivalent.

Example 3B

Using the Least-Square Approach to Calculate the Best-Fit Mass Coefficients and the Residuals at Each Locus

[0116] The method used to calculate the best-fit mass coefficients and the residuals at each locus is presented in this section. The allele peak areas at each locus of a mixed DNA/STR sample are a superposition of both contributors' allele peaks, weighted by their corresponding mass proportions. For example, for a locus with three alleles, the relationship between the genotypes of the two contributors and the allele peak area measured can be expressed as a linear equation:

[0117]

[\begin{matrix} 1 & 0 \\ 0 & 1 \\ 1 & 1 \end{matrix}] * [\begin{matrix} Mass 1 \\ Mass 2 \end{matrix}] = Mass 1 [\begin{matrix} 1 \\ 0 \\ 1 \end{matrix}] + Mass 2 [\begin{matrix} 0 \\ 1 \\ 1 \end{matrix}] = α [\begin{matrix} Allele 1 & peak data \\ Allele 2 & peak data \\ Allele 3 & peak data \end{matrix}]

A * x = b

where ‘α’ is a proportionality constant. For the locus with three or four alleles, A is a three by two, or four by two matrix respectively, and the multivariable linear equation of Ax=b would generally be inconsistent, with more equations than unknowns and no exact solution would exist, meaning no solution can be found such that Ax would equal to b exactly.

[0118] However, the least-square solution of Ax=b can be found, where x_lsis a two-element vector composed of the relative mass coefficients for the two contributors to the DNA mixture, such that A times x_lswould be ‘closest’ to the given vector ‘b’, of all possible x vectors. Computationally, the least-square solution, {circumflex over (x)} to the inconsistent Ax=b, is given by {circumflex over (x)}=A⁺b, where A⁺ denotes the pseudoinverse of A (16). The solution {circumflex over (x)} is such that (A{circumflex over (x)}) will give a fitted allele peak vector {circumflex over (b)}, which is not equal to b exactly, but would be ‘closest’ to ‘b’. The difference between ‘b’ and {circumflex over (b)} is called the residual vector. The magnitude of the residual, representing how close {circumflex over (b)} is to b is
E²=∥(b−A{circumflex over (x)})∥²
where ∥·∥ stands for the norm of the error vector. Thus, for each locus, each of the possible genotype matrices is fitted to the given allele peak data vector, b, and the associated residual, E², is calculated and compared against each other. The genotype pair that gives the smallest fitting error is ranked the highest as the most likely genotype profiles for the two contributors, at this locus, provided the relative allele peak data is consistent with the underlying relative DNA masses in the mixture.

Example 3C

Finding the Most Likely Genotype Combination for Each Locus

[0119] If the proportionality between the DNA mass fractions and the associated allele peak areas is preserved, then it is expected that the genotype combination that gives the smallest fitting error residual, if a unique minimum exists, corresponds to the true genotype profiles of the two contributors. Furthermore, by comparing the smallest residual to the next smallest one, one can estimate the confidence with which one can claim the genotype with the smallest fitting residual being the correct one. If the next smallest residual is very much bigger than that of the smallest one, then one can claim with more confidence that the genotype with the smallest residual is the correct genotype combination pattern at that locus.

[0120] For a locus with two alleles, there are four possible genotype combinations for the two contributors, as shown in Table 2. The 2-allele case always needs special consideration, because the matrices A of three of the four possible genotype combination cases are full rank. This means that an exact solution exists in solving for x in Ax=b. Therefore, a least-square solution is not involved. Consequently, this also means that the computed mass coefficient vector, x, gives an exact fit to the allele peak area vector and the fitting error is zero. However, this does not necessarily imply that one of these three genotype combination cases is the correct one. It is just that the mathematics involved in solving the problem is such that an exact fit is allowed. Therefore, it is the associated mass ratio that has to be used to determine which genotype combination case is more consistent with the best-fit mass ratios identified from other loci already processed. Two situations exist in the 2-allele case:

[0121] If the genotypes of the two individuals are different, as in Cases 1, 2, and 3 shown in Table 2, then the matrix A is full rank, meaning the two columns of A are independent.

[0122] Therefore, an exact solution x for the equation Ax=b can be obtained from x=A⁺b; in this case, A⁻¹exists and is equal to A⁺. As a result, the errors calculated from these three cases are all zero. For this reason, in this case comparing the errors does not help the caseworkers in selecting the best-fit genotype assignment for a two-allele locus. The mass ratio is more indicative of which case is the best-fit case.

[0123] If the genotypes for both contributors are the same and each individual is heterozygous at this locus, as in Case 4 shown in Table 2, then regardless of the mixture DNA's mass ratio, the fitted two-allele peak areas at this locus will always be the same. The matrix A that represents this genotype assignment is

[0124]

[\begin{matrix} 1 & 1 \\ 1 & 1 \end{matrix}] .

The two vectors of this matrix are identical, and the rank is 1. Therefore, an exact solution for the equation Ax=b can be found only when the two elements in b vector are also the same, meaning the two allele peak areas are equal to each other. Since there is always uncertainty associated with the allele peak data measurement, they are seldom equal exactly. As a result, no exact solution would be obtained, and the fitting error for this case would always be greater than 0, thus greater than those for the other genotype cases. In addition, the computed fitted mass ratio would always be 1:1, but in most cases, it would be wrong. The real mass ratio can be anything, but cannot be estimated from the least-square fit solution. Therefore, when the peak areas of the two alleles are indeed close, then the possibility exists that Case 4 is the applicable genotype combination case, regardless of the value of the fitting error and its unestimatable mass ratio.

[0125] From the above analysis, one can conclude that for loci with only two alleles, comparing the errors associated with each possible genotype combination choice does not help the caseworkers to decide the correct genotypes for the two contributors. The best-fit genotype combination should be the one with a mass ratio consistent with those estimated for other loci with three or four alleles. However, if the ratio of the two allele peak areas

is close to 1:1, the genotype combination Case 4 represented by the matrix

[0126]

[\begin{matrix} 1 & 1 \\ 1 & 1 \end{matrix}]

should then be considered possible, even though the error for this case is greater than those for the other cases.

[0127] In general, the method to identify the most likely genotype combinations at each locus in LSD, is summarized as follows. FIG. 10 provides a detailed flow diagram of the LSD and identification process. (25) For a locus with three or four alleles, the genotype combination with the smallest fitting residual should first be chosen as the most likely one. The smallest fitting residual may not be chosen if it clusters with the second smallest fitting residual. The smallest fitting residual clusters with the second smallest fitting residual if the smallest and second smallest fitting residuals are much more similar to each other than the second smallest fitting residual is similar to the remaining fitting residuals. The smallest and the second smallest fitting residuals do not cluster if the second smallest fitting residual is less similar to the smallest fitting residual than it is to the remaining fitting residuals, if the second smallest fitting residual is not similar to either the smallest or the remaining fitting residuals, or if the second smallest fitting residual is not much more similar to the smallest fitting residual than the remaining fitting residuals. Examples 4-9 apply the LSD method and thus provide fitting residuals, or fitting errors, for genotype profiles that can be selected for individuals that contribute DNA to a sample. These Examples also discuss the selection of one of these genotype profiles for the individuals, considering the fitting residual or fitting errors of the genotype profiles. Thus these Examples can be used to guide one of skill in the art as to whether a smallest fitting residual clusters with a second smallest fitting residual in determining a genotype profile.

[0128] Consistency of mass ratios can also be used in the selection of a genotype profile of a locus if the genotype profile at a second locus has been determined. A consistent mass ratio can refer a mass ratio that is nearly identical for each allele combination selected at each locus in a sample. A consistent mass ratio may also be no more than a 1.2-, 1.5-, 1.8-, 2.0-, 3.0-fold difference between the mass ratios of the genotype profiles of the second locus and the selected locus in the sample.

[0129] For a locus with two alleles the flow diagram shown in FIG. 11 may be applied. If the ratio of the two-allele peak areas is not close to 1:1, the genotype combination with the fitted mass ratio consistent with those for other loci should be selected as the most likely one among the first three cases shown in Table 2. If the ratio of the two allele peak areas is close to 1:1 and the approximate mass ratio of the DNA mixture, derived from other loci, is not close to 1:1, the genotype combination in Case 4 shown in Table 2 {[A, B][A, B]} should then be considered as the only possibility, even though the error for this case is greater than those for the other cases. If the ratio of the two allele peak areas is close to 1:1 and the approximate mass ratio of the DNA mixture is also close to 1:1, then there are two possible genotype combinations {[A, A][B, B]} and {[A, B][A, B]}, both of which can fit equally well to the measured allele peak areas.

Example 3D

Reviewing the Ranking Result and Forming the Composite Profiles

[0130] From pooling together the most likely genotype combinations at all loci, the best-fit genotype profiles can be established for the two source contributors. Since the best-fit mass ratio varies somewhat from locus to locus, an approximate mass ratio of the DNA mixture sample can be obtained by averaging across the individual mass ratios from each locus. It is also possible to consider a range of mass ratios for a majority of the three or four allele loci as a cluster of mass ratios.

[0131] It is also possible to compare the genotype profiles selected with a reference profile to validate the genotype profile choices made, and to further resolve ambiguities may exist when more than one combination at a locus has not been ruled out.

Example 4

Application of Least Square Approach (LSD) Principles Resolves a Mixed Forensic DNA Sample into Each Contributor's Genotype

[0132] LSD was used to successfully resolve a DNA mixture obtained from a rape case into the genotype profile of the victim and the suspect. Examples 4A, B, and C illustrate the application of the LSD approach to resolve the DNA sample into each contributor' genotype profile at each of a three-allele (VWA), four-allele (D21S11), and two-allele locus (D7S820). FIG. 12 shows the output of a Genescan software program used to detect the alleles of the victim and suspect at nine loci (19). The alleles and allele peaks are seen on the data output. The peak areas obtained from the Genescan output and the known genotypes of the victim and the suspect are listed in Table 4. The output of the web-LSD analysis of the allele peak areas is shown in FIGS. 13A-C. The web-LSD was based on previous Visual Basic and Matlab implementations of the software to support LSD (20).

[0133]

	TABLE 4

	Alleles in the		Genotypes
Locus	Mixture	Peak Areas	Victim	Suspect

D3S1358	15	1242	15, 17	16, 17
	16	657
	17	1546
VWA	16	616	17, 17	16, 18
	17	2021
	18	425
FGA	22	850	22, 24	23, 25
	23	468
	24	681
	25	315
D8S1179	12	1842	12, 12	13, 16
	13	490
	16	461
D21S11	30	842	30, 32.2	30.2, 31.2
	30.2	490
	31.2	509
	32.2	804
D18S51	12	440	13, 13	12, 15
	13	1503
	15	387
D5S818	7	486	13, 13	7, 12
	12	512
	13	1886
D13S317	8	734	8, 9	9, 11
	9	1068
	11	299
D7S820	10	614	11, 11	10, 10
	11	1169

Example 4A

Application of LSD Analysis at a Three-Allele Locus to Predict the Profile of the Two Contributors

[0134] As shown in Table 4, locus VWA has three alleles. The allele peak data at this locus are first normalized and represented with a vector ‘b’:

[0135]

b = [\begin{matrix} 616 / 425 \\ 2021 / 425 \\ 425 / 425 \end{matrix}] = [\begin{matrix} 1.45 \\ 4.76 \\ 1 \end{matrix}] .

[0136] All six possible genotype combinations for this 3-allele locus and their corresponding matrix ‘ A’ at this locus are listed in Table 5.

[0137]

TABLE 5

	Genotype Combination		Pseudoinverse of the
Case	Person 1	Person 2	Matrix (A)	Matrix (A⁺)

1	16, 16	17, 18	[ 2 0 0 1 0 1 ]	1 2 ⁡ [ 1 0 0 0 1 1 ]

2	17, 17	16, 18	[ 0 1 2 0 0 1 ]	1 2 ⁡ [ 0 1 0 1 0 1 ]

3	18, 18	16, 17	[ 0 1 0 0 2 1 ]	1 2 ⁡ [ 0 0 1 1 1 0 ]

4	16, 17	17, 18	[ 1 0 1 1 0 1 ]	1 3 ⁡ [ 2 1 - 1 - 1 1 2 ]

5	16, 17	16, 18	[ 1 1 1 0 0 1 ]	1 3 ⁡ [ 1 2 - 1 1 - 1 2 ]

6	16, 18	17, 18	[ 1 0 0 1 1 1 ]	1 3 ⁡ [ 2 - 1 1 - 1 2 1 ]

[0138] For each possible genotype combination, the least-square solution {circumflex over (x)} for the equation Ax=b is solved by multiplying the corresponding pseudoinverse of A, (i.e. A⁺) to the vector ‘b’. The error residual was calculated. Case 1 is used as an example to show how the residual error is calculated. For Case 1, the corresponding A matrix is:

[0139]

[\begin{matrix} 2 & 0 \\ 0 & 1 \\ 0 & 1 \end{matrix}] .

The linear equation for this case is:

[0140]

Ax = b \Rightarrow [\begin{matrix} 2 & 0 \\ 0 & 1 \\ 0 & 1 \end{matrix}] x = [\begin{matrix} 1.45 \\ 4.76 \\ 1 \end{matrix}],

and the least-square solution {circumflex over (x)} is decided as follows:

[0141]

\hat{x} = A^{+} b = \frac{1}{2} [\begin{matrix} 1 & 0 & 0 \\ 0 & 1 & 1 \end{matrix}] \cdot [\begin{matrix} 1.45 \\ 4.76 \\ 1 \end{matrix}] = [\begin{matrix} 0.73 \\ 2.88 \end{matrix}],

where A⁺ is obtained from Table 2. The residual for the fit is calculated as:

[0142]

E^{2} = { (b - A \hat{x}) }^{2} = { ([\begin{matrix} 1.45 \\ 4.76 \\ 1 \end{matrix}] - [\begin{matrix} 2 & 0 \\ 0 & 1 \\ 0 & 1 \end{matrix}] \cdot [\begin{matrix} 0.73 \\ 2.88 \end{matrix}]) }^{2} = { ([\begin{matrix} - 0.01 \\ 1.88 \\ - 1.88 \end{matrix}]) }^{2} = {(- 0.01)}^{2} + {1.88}^{2} + {(- 1.88)}^{2} = 7.0689

[0143] The normalized mass ratio vector is:

[0144]

[\begin{matrix} 0.73 / 0.73 \\ 2.88 / 0.73 \end{matrix}] = [\begin{matrix} 1 \\ 3.95 \end{matrix}],

which is the normalized least-square solution {circumflex over (x)} reflecting the approximate mass ratio in the mixture sample for that locus. The least-square solution {circumflex over (x)} and the associated error residual for other cases were also calculated in the same manner as that for Case 1, and results are shown in Table 6.

[0145]

TABLE 6

	Genotype Combinations		Ratio of	Mass Ratio
Cases	Person 1	Person 2	Matrix A	Errors	Calculated

1	16, 16	17, 18	[ 2 0 0 1 0 1 ]	69.82	1:3.95

2	16, 18	17, 17	[ 0 1 2 0 0 1 ]	1	1:1.94

3	18, 18	16, 17	[ 0 1 0 0 2 1 ]	54.11	1:6.20

4	17, 18	16, 17	[ 1 0 1 1 0 1 ]	17.55	1:1.25

5	16, 18	16, 17	[ 1 1 1 0 0 1 ]	61.20	1:−7.63

6	16, 18	17, 18	[ 1 0 0 1 1 1 ]	89.42	1:−10.58

[0146] The first column of the table is the case number for each genotype combination; all possible genotype combination cases are listed in the second and third columns; the fourth column of the table is the corresponding genotype matrix; the fifth column is the ratio of the calculated errors, with the smallest error normalized to one the normalized mass ratios calculated from the fitted {circumflex over (x)} are listed in the sixth column.

[0147] FIG. 14 shows the error residual for each of the possible genotype combination cases. Therefore, all the genotype combinations for locus VWA can be ranked according to the magnitudes of the corresponding residual errors. Case 2 clearly has the smallest fitting error. The next smallest error, that of Case 4, has an error 17 times as large as that for Case 2, indicating genotype combination of Case 2 is clearly the best-fit one if the relative allele peak area proportion is credible. Table 7 is the ranking result for locus VWA with three alleles.

[0148]

TABLE 7

	Genotype Combinations		Ratio of	Mass Ratio
Rank	Person 1	Person 2	Matrix A	Errors	Calculated

1	16, 18	17, 17	[ 0 1 2 0 0 1 ]	1	1:1.94

2	17, 18	16, 17	[ 1 0 1 1 0 1 ]	17.55	1:1.25

3	18, 18	16, 17	[ 0 1 0 0 2 1 ]	54.11	1:6.20

4	16, 18	16, 17	[ 1 1 1 0 0 1 ]	61.20	1:−7.63

5	16, 16	17, 18	[ 2 0 0 1 0 1 ]	69.82	1:3.95

6	16, 18	17, 18	[ 1 0 0 1 1 1 ]	89.42	1:−10.58

[0149] From the ranking result, one can see that the fitting error of the first ranked genotype combination is clearly much smaller than the next smallest one; therefore, the genotype combination of {[16, 18], [17, 17]} is identified as the best-fit for this locus, and the mix is at the approximate mass ratio of 1:1.94. The genotype combination with a negative number in the calculated mass coefficients can be excluded at once because the mass coefficients must be positive.

Example 4B

Application of LSD Analysis at a Four-Allele Locus to Predict the Profile of the Two Contributors

[0150] The allele peak data at locus D21S11 shown in Table 4 is used to illustrate how a 4-allele pattern at a locus is resolved into its most likely components using LSD. First, all possible genotype assignments for this locus are listed in Table 8.

[0151]

TABLE 8

	Genotype Combination		Pseudoinverse of the
Case	Person 1	Person 2	Matrix (A)	Matrix (A⁺)

1	30, 30.2	31.2, 32.2	[ 1 0 1 0 0 1 0 1 ]	1 2 ⁡ [ 1 1 0 0 0 0 1 1 ]

2	30, 31.2	30.2, 32.2	[ 1 0 0 1 1 0 0 1 ]	1 2 ⁡ [ 1 0 1 0 0 1 0 1 ]

3	30, 32.2	30.2, 31.2	[ 1 0 0 1 0 1 1 0 ]	1 2 ⁡ [ 1 0 0 1 0 1 1 0 ]

[0152] For each possible genotype combination, the fitting error and the best-fit mass coefficients are calculated. FIG. 15 shows the relative error of each possible genotype combination when normalized to the smallest error (Case 3). The LSD ranking result for this locus is shown in Table 9.

[0153]

TABLE 9

	Genotype Combinations		Ratio of	Mass Ratio
Rank	Person 1	Person 2	Matrix A	Errors	Calculated

1	30.2, 31.2	30, 32.2	[ 1 0 0 1 0 1 1 0 ]	1	1:1.65

2	30.2, 32.2	30, 31.2	[ 1 0 0 1 1 0 0 1 ]	116.06	1:1.04

3	31.2, 32.2	30, 30.2	[ 1 0 1 0 0 1 0 1 ]	116.86	1:1.01

[0154] The top ranked genotype combination of {[30.2, 31.2][30, 32.2]} is considered to be the most likely one with the approximate mass ratio of 1:1.65. Note that the best-fit mass ratio of 1:1.65 for this locus is consistent with the best-fit mass ratio of 1:1.94 for the 3-allele locus VWA, derived in the previous section.

Example 4C

Application of LSD Analysis at a Two-Allele Locus to Predict the Profile of the Two Contributors

[0155] If only two alleles are present at a locus, then there are four possible genotype combinations for the two contributors, as shown in Table 10.

[0156]

TABLE 10

	Genotype Combination		Pseudoinverse of
Case	Person 1	Person 2	Matrix (A)	the Matrix (A⁺)

1	A, A	B, B	[ 2 0 0 2 ]	1 2 ⁡ [ 1 0 0 1 ]

2	A, B	A, A	[ 1 2 1 0 ]	1 2 ⁡ [ 0 2 1 - 1 ]

3	A, B	B, B	[ 1 0 1 2 ]	1 2 ⁡ [ 2 0 - 1 1 ]

4	A, B	A, B	[ 1 1 1 1 ]	1 4 ⁡ [ 1 1 1 1 ]

[0157] All possible genotype assignments for this locus are listed in Table 4.11 based on the allele contributions of the individuals at locus D7S820 using the data shown in Table 4.

[0158]

TABLE 11

	Genotype Combination		Pseudoinverse of the
Case	Person 1	Person 2	Matrix (A)	Matrix (A⁺)

1	10, 10	11, 11	[ 2 0 0 2 ]	1 2 ⁡ [ 1 0 0 1 ]

2	10, 11	10, 10	[ 1 2 1 0 ]	1 2 ⁡ [ 0 2 1 - 1 ]

3	10, 11	11, 11	[ 1 0 1 2 ]	1 2 ⁡ [ 2 0 - 1 1 ]

4	10, 11	10, 11	[ 1 1 1 1 ]	1 4 ⁡ [ 1 1 1 1 ]

[0159] At this locus, the errors of all possible genotype combinations are calculated. The calculated errors and the ranking result for this locus are shown in Table 12.

[0160]

TABLE 12

[0161] Since the peak area ratio of the two alleles (614:1169) at locus D7S820 is not close to 1:1, one does not need to consider Case 4 with the genotype combination of {[10, 11][10, 11]} as a possibility. Recall the best-fit mass ratios from loci VWA and D21S11 are 1:1.94 and 1:1.65, respectively. Therefore, either Case 1 or Case 3 is a possibility with the fitted mass ratio of 1:1.9 and 1:2.21, respectively. These two genotype combinations are {[10, 10][11, 11]} and {[11, 11][10, 11]}. The genotype of the victim, if available, can definitely help the caseworkers to select the more appropriate genotype combination case from the two alternative choices. If more than one case yields a mass ratio consistent with those from other loci, then DNA for this locus cannot be resolved unambiguously, which means LSD suggests more than one genotype combination, both of which are equally likely to be the true case. The only way to ascertain the choice is to pick out the genotype combination case, in which the known victim's genotype or the reference genotype is one of the genotype profiles in the combination. It turns out the combination of {[10, 10][11, 11]} is the pick of the choice because the victim's (person 2) profile at this locus is [11, 11].

[0162] The other six loci of Table 4 were analyzed with the same LSD interpretation approach. Compared with the true genotypes of the two contributors, LSD give the correct genotype profiles at all 3-allele and 4-allele loci, for this example. For those 2-allele loci, LSD yielded two possible choices for each locus. When the genotype of the victim at this locus was brought in, the better choice was identified. Therefore, in this case, LSD results at all nine loci are consistent with the genotypes of the victim and the suspect.

Example 5

The LSD Approach Correctly Predicts the Genotype Profile of Two Contributors in a Simulated DNA Sample with Two Contributors

[0163] Allele peak data of DNA mixture samples were simulated at six loci. The loci contained two, three, or four alleles and mass ratios of 1:1, 1:2, and 1:5. The sets of simulated data are shown in Table 13.

[0164]

TABLE 13

		Genotype	Simulated Allele Peak
	Alleles in	Combination	Data
Locus	the Mixture	Person 1	Person 2	1:5	1:2	1:1

VWA	15	15, 15	16, 19	1302	1506	1472
	16			3369	1369	730
	19			3446	1348	769
TH01	6	6, 7	6, 9.3	5620	5008	5979
	7			807	1875	3057
	9.3			5095	3155	3122
D8S1179	11	11, 13	11, 13	5519	5284	4339
	13			4913	5733	4922
FGA	22	22, 22	22, 25	9019	4675	8596
	25			6240	2232	2444
D21S11	28	28, 28	30.2, 30.2	958	1098	1001
	30.2			5234	1941	873
D18S51	16	16, 20	17, 18	804	731	975
	17			5202	1806	882
	18			4212	1640	899
	20			864	913	873

[0165] First, at each locus, the genotype combination of the two contributors is coded into a matrix ‘A.’ In accordance to the mass ratio assumed for each set of data, the corresponding theoretical allele peak data vector ‘b’ at this locus was then calculated from the equation Ax=b, where the vector contains the mass ratio coefficients for the two contributors. In real cases, some noise or peak area imbalances are always associated with the measurements of allele peak areas. Therefore, in the simulated data, a 10% Gaussian noise (random Gaussian noise with mean 0 and standard deviation of 10% of the nominal peak area) is added to the theoretical allele peak values.

Example 5A

A Simulated DNA Mixture with a Mass Ratio of 1:5

[0166] The simulated DNA mixture shown in Table 13 for the mass ratio of 1:5 was analyzed by the LSD method. FIGS. 16A and 16B show the detailed web-LSD output for this set of data. Table 14 is the LSD suggested genotype resolution result of analyzing the simulated data for a mass ratio of 5:1.

[0167]

TABLE 14

	Alleles	Peak Area	LSD Suggested Genotype Resolution Result
	in the	Ratio		Mass Ratio
Locus	Mixture	Measured	Person 1	Person 2	Calculated	Remarks

VWA	15, 16, 19	1:	15, 15	16, 19	1:5.23	Correct
		2.59:
		2.65
TH01	6, 7, 9.3	6.96:	6, 7	6, 9.3	1:7.01	Correct
		1:
		6.31
D8S1179	11, 13	1.12:	11, 13	11, 13	Indeterminate	See Table
		1				15 for
						details
FGA	22, 25	1.45:	22, 22	22, 25	1:4.49	Correct
		1
D21S11	28, 30.2	1:	28, 28	30.2, 30.2	1:5.46	Correct
		5.46
D18S51	16, 17, 18,	1:	16, 20	17, 18	1:5.64	Correct
	20	6.47:
		5.24:
		1.07

Note: ‘correct’ in the table means the top-ranked genotype combination selected by LSD corresponds to the true genotype profiles of the two contributors.

[0168] For loci VWA, TH01, FGA, D21S11, and D18S51, the top-ranked genotype combinations suggested by LSD correspond to the correct genotype profiles of the two contributors. At these five loci, the fitted mass ratios are reviewed to be more or less consistent with each other ranging from 1:4.49 to 1:7.01. The average of the five mass ratios is 1:5.57, which is close to the true mass ratio of 1:5 in this simulated DNA mixture sample.

[0169] The remaining locus of D8S1179 has only 2 alleles, and LSD suggested results for a 2-allele locus always needs special review and interpretation. The ranking result for all possible genotype combinations for this locus is presented in Table 15.

[0170]

TABLE 15

[0171] Three genotype combinations all fit the given allele peak data exactly with zero errors. However, their corresponding best-fit mass ratios are 1:1.1, 1: −18.2, and 1:16.2; all are inconsistent with those for other loci. For the genotype combination of {[11, 13][11, 13]}, if it were true, then regardless of the mass ratio in the DNA mixture, the ratio of the allele peak data would always be close to 1:1, and it is (1.12:1). As a result, the true mass ratio would be unknown at this locus. The best-fit mass ratio in this case, given by LSD usually is not the correct one, due to the ‘A’ matrix for this case being underdetermined (meaning not all rows or columns of ‘A’ are independent), and mathematically an infinite set of solutions exist and LSD would give the one with the smallest length, the normalized result would always be 1:1. Therefore, for the case of 2-allele loci, when the other three genotype combinations all give a best-fit mass ratio that is inconsistent with those for other loci, the fourth case is the preferred choice, especially when the ratio of the two allele peaks is close to 1:1. As a result, we choose this genotype combination to be the most likely one. It turns out that it is the correct one.

Example 5B

A Simulated DNA Mixture with a Ratio of 1:2

[0172] Table 13, above, also contains the data for the simulated DNA mixture with a mass ratio of 1:2. This data was also input into the LSD method for analysis. FIGS. 17A and B show the detailed web-LSD output for this data set. Table 16 is the resulting LSD suggested genotype resolution obtained for this simulated data set.

[0173]

TABLE 16

[0174] LSD guided genotype combination pair at each locus of VWA, TH01, and D18S51 can be picked out by first ranking the least-square fitting residuals for all genotype combinations and then selecting the top-ranked one with the smallest fitting errors.

[0175] At the two-allele locus D8S1179, the genotype combination of {[11, 13][11, 13]} is selected as the most fitting one, because the ratio of the two allele peak areas is close to 1:1 (1:1.09) and the other patterns' fitted mass ratios are not consistent with those for other loci.

[0176] For locus FGA, the ranking result for all possible genotype combinations is presented in Table 17. For a 2-allele locus, when the mass ratio of the DNA is close to 1:2, and the ratio of the two allele peak areas is also close to 2:1 (or 1:2), two genotype combinations of {[B, B][A, A]} and {[A, A][A, B]} will both generate two allele peaks with the ratio of the peak A to peak B area being close to 2:1. Therefore, LSD guided results generate two best-fit genotype combinations for each of the two loci. This is a limitation of LSD imposed by the underlying mathematics. In this case, one can make the decision as to which one is the best choice, after the genotype of the victim is brought in for reference. The analysis for locus D21S11 is similar. In Table 16, the true genotypes of the two contributors are indicated as shaded cells.

[0177]

TABLE 17

Example 5C

A Simulated DNA Mixture with a Ratio of 1:1

[0178] Table 13, above, also contains the data for the simulated DNA mixture with a mass ratio of 1:1. This data was input into the LSD method for analysis. FIGS. 18A and B show the detailed web-LSD output for this data set. Table 18 shows the suggested genotype resolution using the output results of the LSD method.

[0179]

TABLE 18

[0180] Mixtures with mass ratios close to 1:1 are difficult to resolve, because many genotype combinations would fit equally well to the given peak data, and the slightest peak imbalance would change the best-fit genotype combination pair. For DNA mixture data with the mass ratio close to 1:1, one cannot differentiate the major and minor contributors clearly even at a 4-allele locus. As a result, the LSD method would yield multiple genotype combinations, all with comparable fit. Therefore, the LSD suggested genotype resolution result is not as clear compared with those of other sets of DNA mixture data with mass ratios other than 1:1. From Table 18, it can be seen that LSD generated at least two possible choices for each of the six loci; the true cases are indicated with shaded cells. However, when a reference profile (such as the victim's) is brought in, the right combination pair can usually be selected from the set of possible pairs. It is crucial to have a reference profile available when the underlying mass ratio is close to 1:1.

[0181] For loci VWA and TH01, each locus has three detected alleles with the ratio of the allele peak areas being close to 2:1:1. Two genotype combinations of {[A, A][B, C]} and {[A, B][A, C]} are possible: both would fit equally well to the peak areas ratio if the mass ratio of the DNA mixture is close to 1:1. Since there is no differentiation between the major and minor contributor, LSD suggests four possible allele assignment patterns for each of the two loci of VWA and TH01, as shown in Table 18, not being able to consistently distinguish between person 1 and person 2's profiles.

[0182] For loci D8S1179 and D21 S11, each locus has two alleles with the ratio of the allele peak data close to 1:1. Since the mass ratio in the mixed DNA sample is also close to 1:1, there are three possible genotype combinations, {[A, B][A, B]}, {[A, A][B, B]}, and {[B, B][A, A]}, which all give comparable allele peak signals. As a result, all three possibilities need to be considered. If the victim's profile is known, then the correct choice can usually be identified.

[0183] Since the peak area of the two alleles present at locus FGA are not comparable, the genotype combination of {[22, 25][22, 25]} can be excluded directly. By comparing the mass ratios calculated from all possible genotype combinations at this locus with those estimated from other loci, the allele assignment pattern of {[22, 25][22, 22]} was selected as the best-fit one.

[0184] For locus D18S51 with four alleles all with comparable peak areas, all six possible genotype combinations are ranked equally because the mass ratio in this DNA mixture sample is around 1:1. The true one is shown in the shaded cells. Since there is always some noise or imbalance associated with the allele peak quantitation, in this case, an LSD-guided analysis would be very sensitive to the relative amount of noise present in the respective allele signals. Therefore, a reference genotype becomes indispensable in assigning a decisive genotype combination pair to the two contributors.

Example 6

Using Data Reported in the Scientific Literature LSD Correctly Predicts the Genotype Profile of Each Individual that Contributed to a Mixed DNA Sample

[0185] LSD was used to analyze raw data obtained from processed DNA mixtures in the scientific literature. Since both contributors' genotypes in the mixture of these reports were known, they were used to evaluate the ability of LSD to successfully predict genotypes by comparing the LSD-suggested genotype resolution with the true genotypes of both contributors.

Example 6A

[0186] DNA mixture data obtained from the report “Taking Account of Peak Areas When Interpreting Mixed DNA Profiles,” by Clayton et al. (3). Table 19 summarizes the data provided by Clayton et al. for six loci of a mixed DNA sample

[0187]

	TABLE 19

	Alleles in the	Genotype Combination	Allele Peak
Locus	Mixture	Person 1	Person 2	Areas

VWA	15	17, 19	15, 16	1247
	16			1193
	17			2279
	19			2000
TH01	5	5, 7	7, 7	5735
	7			10769
D8S1179	13	13, 13	14, 15	3234
	14			752
	15			894
FGA	22	23, 23	22, 23	534
	23			2792
D21S11	61	65, 70	61, 77	373
	65			590
	70			615
	77			356
D18S51	14	16, 18	14, 15	1339
	15			1465
	16			2895
	18			2288

[0188] These data were input into and analyzed by the LSD method. The results of the LSD analysis are shown in FIGS. 19A and B. The results of the data analysis are shown in Table 20.

[0189]

TABLE 20

[0190] By comparing the LSD suggested result with the true genotypes of both contributors, one can see that among the six loci, the top-ranked genotype combinations selected are consistent with that of the correct one at five loci (all but TH01). From the result at these five loci, one can see that the approximate mass ratio of the DNA mixture is around 1:2. Locus TH01 has only two alleles and post-LSD review is always required. The LSD analysis yields two possible allele assignment patterns (both of which have zero fitting residuals) with the calculated mass ratios consistent with those estimated from the other five loci. If the DNA analyst does not know the genotype of either contributor, then she cannot make the decision as to which one to choose.

Example 6B

[0191] DNA mixture data obtained from the report “Taking Account of Peak Areas When Interpreting Mixed DNA Profiles,” by I. Evett et al. (21). All available information about this set of DNA mixture data is presented in Table 21

[0192]

	TABLE 21

		Genotype
	Alleles	Combination	Allele
Locus	in the Mixture	Suspect	Victim	Peak Areas

TH01	8	8, 9.3	9.3, 9.3	17441
	9.3			22368
D21S11	59	67, 70	59, 65	1226
	65			1434
	67			8816
	70			8894
D18S51	13	13, 13	16, 17	38985
	16			1914
	17			1991
D8S1179	10	10, 14	10, 11	6416
	11			383
	14			5659
VWA	16	16, 18	17, 19	4669
	17			931
	18			4724
	19			188
FGA	21	21, 22	21, 23	16099
	22			10538
	23			1014

[0193] The data were analyzed by the LSD method. FIGS. 20A and B show the output of the web-LSD analysis of the data. The top-ranked genotype profiles obtained using the LSD analysis are shown in Table 22.

[0194]

TABLE 22

	Alleles	Peak Area	LSD Suggested Genotype Resolution Result
	in the	Ratio		Mass Ratio
Locus	Mixture	Measured	Person 1	Person 2	Calculated	Remarks

TH01	8	1:	9.3, 9.3	8, 9.3	1:7.08	With
	9.3	1.28				confidence
D21S11	59	1:	59, 65	67, 70	1:6.66	With
	65	1.17:				confidence
	67	7.19:
	70	7.25
D18S51	13	20.37:	16, 17	13, 13	1:9.98	With
	16	1:				confidence
	17	1.04
D8S1179	10	16.75:	10, 11	10, 14	1:11.39	With
	11	1:				confidence
	14	1.476
VWA	16	24.84:	17, 19	16, 18	1:8.39	With
	17	4.95:				confidence
	18	25.13:
	19	1
FGA	21	15.88:	21, 23	21, 22	1:4.76	With
	22	10.39:				confidence
	23	1

[0195] Although the six best-fit mass ratios calculated from the six loci are not exactly consistent, ranging from 1:4.8 (FGA) to 1:11.4 (D8S1179), this mixed DNA sample can still be resolved confidently. At each of the six loci, the genotypes of Person 1 and Person 2 correspond to the true genotypes of victim and the suspect, respectively. As a result of the noise and inconsistent PCR amplification among the six loci, there is some difference between the calculated average mass ratio of 1:8.04 and the known one prior to amplification in this DNA mixture of 1:10. However, such a difference does not affect the resolution result of the genotype profiles for the two contributors.

[0196] From the above interpretation, one can see that both of these two sets of DNA mixture data from literature reports were correctly resolved using LSD by selecting the top-ranked genotype combination with consistent mass ratio at each locus. No information about the genotype of any contributor and the mass ratio in the mixture data is necessary for LSD to resolve them. Such information was only used to verify the LSD suggested genotype resolution results.

Example 7

LSD Correctly Identifies the Genotype Profile of Contributors to DNA Mixture Samples Obtained from the Acadiana Crime Lab

[0197] To further test the effectiveness of LSD, DNA mixture samples from real forensic cases were obtained and analyzed. In each instance, the mixed DNA profiles were successfully resolved using LSD.

[0198] For each mixed DNA sample the allele peaks were scanned and measured by an ABI Prism 310 Capillary Electrophoresis Gene Sequencer using Genescan and Genotyper software. The allele peak areas were measured at all 13 CODIS core loci. After applying LSD to each of the sets of data, a composite profile for each data sample was obtained by identifying the top-ranked genotype combination for each locus and choosing those with consistent mass ratios.

[0199] DNA analysts knew the genotypes of both contributors in each case. The LSD-predicted genotype profile for each contributor in the mixture was verified as correct using the known genotype profiles. The results of using LSD to resolve three of these DNA mixtures is described in Examples 7A, B, and C.

Example 7A

LSD Accurately Predicts the Genotype Profiles of Each Contributor in Acadiana DNA Mixture Number 5

[0200] Table 23 provides the allele peak data measured from thirteen loci in DNA mixture number 5. The known genotype profile of each contributor is also provided in this table.

[0201]

TABLE 23

Alleles in	Allele Peak	Contributors'
the	Heights	Genotypes
Locus	Mixture	Measured	Suspect	Victim

D3S1358	14	616	14, 16	17, 19
	16	688
	17	3592
	19	3479
VWA	14	3483	14, 16	14, 16
	16	3534
FGA	22	5503	22, 22	22, 22
D8S1179	13	3111	13, 14	13, 14
	14	3175
D21S11	28	3166	29, 29	28, 30
	29	489
	30	2964
D18S51	13	2383	14, 18	13, 15
	14	380
	15	2041
	18	271
D5S818	11	538	11, 11	12, 13
	12	2448
	13	2240
D13S317	11	309	11, 13	12, 14
	12	1892
	13	366
	14	1734
D7S820	8	1685	8, 12	8, 10
	10	1409
	12	144
D16S539	11	3047	12, 13	11, 14
	12	460
	13	562
	14	2329
TH01	9	2276	9, 9.3	9, 9.3
	9.3	2666
TPOX	8	2556	8, 11	8, 11
	11	2447
CSF1PO	10	1693	11, 12	10, 12
	11	339
	12	2164

[0202] These data were analyzed by the LSD method. The output of the web LSD application is shown in FIGS. 21A-D.

[0203] The LSD predicted genotype profile of each contributor suggested by the LSD output is shown in Table 24.

[0204]

TABLE 24

[0205] Twelve of thirteen total loci are correctly resolved into composite genotype profiles for each contributor LSD. At each of the twelve loci, the genotype profile suggested by LSD corresponds to the true genotype profile of the victim and the suspect.

[0206] At loci D3S1358, D21S11, D18S51, D5S818, D13S317, D7S820, D16S539, and CSF1PO, the genotype profile is easily determined because the top-ranked genotype combination in each case has a small fitting error compared to the second ranked genotype combination, and an associated mass ratio that is consistent with the mass ratio of the other loci, approximately 1:6.26.

[0207] At locus FGA, only one allele is observed. Thus it both the victim and the perpetrator have the genotype of [22, 22] at this locus.

[0208] At each locus of loci VWA, D8S1179, TH01, and TPOX, two alleles are observed with ratios of peak areas close to 1:1. The genotype profile of each contributor is indeterminate by least squares analysis alone. However, the mass ratios of the DNA mixture sample estimated from other loci are not close to 1:1. Because none of the mass ratios for the first three genotype cases at each of these four loci is close to the majority of the best-fit mass ratios of the loci having three or four alleles, the first three cases of allele combinations may be eliminated from consideration, and the only possible genotype combination is the fourth case. At these four loci, the LSD method leads to the conclusion that both individuals have the same allele combinations of {A,B}. This is an example of the need for review of the information provided by the least squares analysis in order to obtain correct results.

[0209] At locus D21S11, the LSD method alludes to two possible choices, because the associated fitting errors are comparable. Detailed LSD ranking result for this locus is shown in Table 25.

[0210]

TABLE 25

[0211] From the ranking result, one can see that both of the top two ranked genotype combinations can fit the measured allele peak data well, and the mass ratio calculated from the second ranked allele assignment (1:7.29) is more consistent with the average (1:6.26) than that calculated from the top ranked one (1:12.54). Therefore, both genotype combinations have to be considered to be the possible one. After this point, the victim's profile is brought in to compare to that of the LSD suggested result. It turned out that the perpetrator has also been identified and typed, and the first choice with person 1 being [29, 29] is the correct one. The victim is found to be LSD's Person 2. Even with the known victim's genotype of [28, 30] at this locus, one still cannot make the correct choice, because both top-ranked genotype combinations have [28, 30] as the genotype for Person 2. In actual application, we suggest that this locus be deleted from the final composite profile for the perpetrator for any follow-up consideration. The cause that the mass ratio associated with the top-ranked genotype combination is much larger than those from other loci. This imbalance in the peak areas is probably the result of imbalanced PCR amplification.

Example 7B

LSD Accurately Predicts Some Genotype Profiles of Contributors in Acadiana DNA Mixture Number 1 Despite Peak Area Saturation in the Data Readout

[0212] Table 26 provides the allele peak data measured from thirteen loci in DNA mixture number 1 and the known genotype profile of each contributor.

[0213]

	TABLE 26

		Contributors'
	Allele Peak	Genotypes
	Alleles in the	Heights	Person 1	Person 1
Locus	Mixture	Measured	(Victim)	(Victim)

D3S1358	14	5786	15, 17	14, 17
	15	1426
	17	5578
VWA	15	5899	16, 19	15, 16
	16	5890
	19	1169
FGA	22	4636	25, 25	22, 24
	24	4765
	25	866
D8S1179	11	6069	13, 15	11, 14
	13	1959
	14	4769
	15	1425
D21S11	28	6479	28, 30.2	28, 32.2
	30.2	1239
	32.2	5717
D18S51	16	1323	16, 16	17, 20
	17	3076
	20	2748
D5S818	11	5354	11, 12	11, 11
	12	1153
D13S317	8	5227	11, 12	8, 12
	11	1136
	12	5208
D7S820	10	5913	10, 12	10, 10
	12	385
D16S539	8	7082	10, 11	8, 10
	10	6995
	11	1300
TH01	6	6388	7, 9.3	6, 7
	7	6370
	9.3	1287
TPOX	8	1393	8, 12	9, 9
	9	7149
	12	1219
CSF1PO	8	729	8, 11	10, 11
	10	4251
	11	4599

[0214] The data provided in Table 26 was analyzed by the LSD method. The output of the web LSD application is shown in FIGS. 22A-D. The LSD predicted genotype profile of each contributor suggested by the LSD method is shown in Table 27.

[0215]

TABLE 27

	Alleles	Peak Area	LSD Suggested Genotype Resolution Result
	in the	Ratio	Person 1	Person 2	Mass Ratio
Locus	Mixture	Measured	(Victim)	(Suspect)	Calculated	Remarks

D3S1385	14	4.06:	15, 15	14, 17	1:7.97	Wrong
	15	1:
	17	3.91
VWA	15	5.05:	19, 19	15, 16	1:10.09	Wrong
	16	5.04:
	19	1
FGA	22	5.35:	25, 25	22, 24	1:10.86	Correct
	24	5.50:
	25	1
D8S1179	11	4.26:	13, 15	11, 14	1:3.20	Correct
	13	1.37:
	14	3.35:
	15	1
D21S11	28	5.23:	28, 30.2	28, 32.2	1:5.15	Correct
	30.2	1:
	32.2	4.61
D18S51	16	1:	16, 16	17, 20	1:4.40	Correct
	17	2.33:
	20	2.08
D5S818	11	4.64:	12, 12	11, 11	1:4.64	Wrong
	12	1
D13S317	8	4.60:	11, 11	8, 12	1:9.19	Wrong
	11	1:
	12	4.58
D7S820	10	15.36:	10, 12	10, 10	1:7.18	Correct
	12	1
D16S539	8	5.45:	11, 11	8, 10	1:10.83	Wrong
	10	5.38:
	11	1
TH01	6	4.96:	9.3, 9.3	6, 7	1:9.91	Wrong
	7	4.95:
	9.3	1
TPOX	8	1.14:	8, 12	9, 9	1:2.74	Correct
	9	5.86:
	12	1
CSF1PO	8	1:	8, 11	10, 11	1:6.85	Correct
	10	5.83:
	11	6.31

[0216] Peak areas over 4,000 units compromise the linear relationship with DNA quantity, due to peak area saturation. Because many allele peak areas measured for this set of data are larger than 4,000 units, they cannot represent the relative mass of such alleles accurately. As a result, only profiles of seven out of a total of thirteen loci correspond to the top-ranked genotype combinations in the result.

[0217] At the other six loci, LSD analysis failed to yield the correct genotype profiles.

[0218] For example: at locus VWA, the true genotypes for the two contributors are [16, 19] and [15, 16], where allele 16 is shared by both contributors and should have a larger peak area than either of the other two alleles. However, the peak areas measured at alleles 15, 16, and 19 are 5899, 5890, and 1169 respectively, where the peak area of allele 16 (5890) is not bigger than that of allele 15 (5899) at this locus. This peak area inversion is the reason that LSD's top choice does not give the correct result. For the same reason, LSD failed to yield the correct allele assignments at the other five loci: D3S1358, D5S818, D13S317, D16S539, and TH01. As for the other seven loci, in spite of allele peak-area saturation, leading to inconsistent best-fit mass ratios, the correct genotype combinations can still be obtained from the top-ranked profiles from the LSD ranking result.

Example 7C

LSD Predicted Genotype Profiles for Each Contributor in Acadiana DNA Mixture Number 2

[0219] Table 28 provides the allele peak data measured from thirteen loci in DNA mixture number 2. The known genotype profile of each contributor is also provided in this table.

[0220]

	TABLE 28

		Contributors'
	Allele Peak	True Genotypes
	Alleles in the	Heights	Person 1	Person 2
Locus	Mixture	Measured	(Victim)	(Suspect)

D3S1358	15	754	16, 18	15, 16
	16	1253
	18	551
VWA	14	489	14, 17	17, 17
	17	1178
FGA	22	757	22, 23	22, 23
	23	626
D8S1179	11	513	11, 12	12, 15
	12	1013
	15	612
D21S11	30	1108	30, 30	30, 31.2
	31.2	332
D18S51	12	150	17, 18	12, 17
	17	415
	18	190
D5S818	10	1164	10, 11	10, 11
	11	948
D13S317	9	339	9, 12	12, 12
	12	957
D7S820	8	279	8, 11	11, 11
	11	463
D16S539	11	506	11, 12	12, 13
	12	765
	13	304
TH01	7	1143	7, 7	7, 9.3
	9.3	374
TPOX	9	809	9, 10	9, 10
	10	645
CSF1PO	10	267	10, 11	11, 11
	11	527

[0221] The data provided in Table 28 was analyzed by the LSD method. The output of the web LSD application is shown in FIGS. 23A-C.

[0222] The LSD predicted genotype profile of each contributor suggested by the LSD method is shown in Table 29.

[0223]

TABLE 29

[0224] LSD guided results for this DNA mixture data suggest at least two genotype combinations for each of the thirteen loci. The calculated mass ratios at several loci are very close to 1:1. Therefore, there is no major and minor contributor, and it becomes difficult to assign which genotype to which person at each locus.

[0225] At six loci, D3 S1358, VWA, D21S11, D13 S317, D165539, and TH01, the best-fit genotype combination at each locus can be selected by comparing the fitting errors and selecting combinations with consistent mass ratios.

[0226] For locus D7S820, the true genotypes of the contributors {[8, 11][11, 11]} gave a best fitted mass ratio of 1:3, far from the best-fit mass ratio obtained at the other loci which are all less than or around 1:2. Therefore, this combination was not selected as a plausible choice. Table 30 shows the detailed LSD fitting for this locus.

[0227]

TABLE 30

[0228] Theoretically, when the mass ratio is close to 1:1, for the genotype combination of {[8, 11][11, 11]}, the ratio of the allele peak data should be more or less equal to 1:3. But the ratio of the allele peak data measured at this locus is 1:1.66, so the genotype combination pair of {[8, 8][11,11]} is selected with the mass ratio of 1:1.66, which is more consistent with the estimated mass ratio of 1:1.

[0229] At loci D8S1179 and D18S51, LSD fitting produced four (out of a total six) possible allele assignment patterns for each locus. The reason is that the ratio of the allele peak data measured at each locus is close to 1:2:1, and the mass ratio in the DNA mixture estimated from several loci is close to 1:1. Two possible genotype combinations can fit equally well to the allele peak data and each genotype combination can have two allele assignment patterns with person 1 and person 2 permuted.

[0230] Therefore, each locus has four genotype combination patterns that need to be taken into consideration. The detailed ranking result at locus D8S1179 is shown in Table 31.

[0231]

TABLE 31

[0232] The errors as well as the mass ratios of the top two ranked genotype combinations are very close to each other. Therefore, it is impossible for one to make a clear decision as to which is overwhelmingly the correct choice, unless the genotype of one contributor is known.

[0233] For loci FGA, D5S818, and TPOX, each locus has only two alleles, and the ratio of the two allele peak data happens to be close to 1:1. Three allele assignment patterns can fit equally well to the allele peak data, and all of them were selected as possible choices for this locus (two of which are permuted arrangement of each other). The caseworkers would not be able to decide which one to choose, unless the genotype of one contributor is available.

[0234] For the 2-allele locus CSF1PO, since the ratio of the two allele peak areas is very close to 1:2, there are two possible allele assignment patterns at this locus: {[A, A][B, B]} and {[B, B][A, B]}, both of which would fit equally well to the measured allele peak data with comparable mass ratios. The approximate mass ratio estimated from other loci in the DNA mixture is around 1:1, which makes it difficult to differentiate the major and minor contributors. Therefore, the permuted allele assignment patterns for person 1 and person 2 of these two cases should also be taken into consideration. As a result, LSD suggests a total of four genotype combination pairs for this locus.

[0235] For this set of DNA mixture data, LSD does not suggest a very clear-cut resolution due to the close mass ratio of 1:1 However, as long as the genotype of one contributor is known, LSD is still useful in yielding the most fitting genotype profile for the other contributor. This is because it is the special combination of the genotypes of the two contributors, as stipulated by the LSD genotype combination pair that minimizes the fitting error, that is formed when fitting to the allele peak data. A genotype profile that includes only those loci that can be determined can be used in this instance.

Example 8

LSD Correctly Identifies the Genotype Profile of Contributors in DNA Mixture Samples from the Texas Department of Public Safety in Corpus Cristi

[0236] To further test the functionality and effectiveness of the LSD method, five additional sets of allele peak data from forensic mixed DNA samples were run using LSD. The DNA mixture data were obtained from swabs in rape cases. The genotype of the female victim is known in each case. Sperm fraction in the mixture sample was separated from the mixture evidence by differential extraction. Therefore, genotype profiles of the male perpetrators are also obtained for all five samples. LSD was applied to the mixture data and the LSD suggested results were compared to the known genotype profiles of the two contributors. Among the five sets of data, four have peak area information at nine loci, and the fifth has information at twelve loci. Two of the five sets of DNA mixture data from Corpus Christi and the corresponding LSD suggested genotype resolution results are discussed in Examples 8A-B.

Example 8A

LSD Predicts the Genotype Profiles of Each Contributor in Texas DNA Mixture Data Number 3

[0237] The electropherogram showing the allele peak data at nine loci for Texas DNA mixture number 3 is shown in FIG. 24. The peak area data extracted from the electropherogram and the known genotype profiles of the female victim and male suspect are shown in Table 32.

[0238] The data provided in Table 32 were analyzed by the LSD method.

[0239]

	TABLE 32

	Alleles in	Allele Peak	Contributors'
	the	Areas	Genotypes
Locus	Mixture	Measured	Female Victim	Male Suspect

D3S1358	15	2935	15, 16	15, 17
	16	1791
	17	999
VWA	16	5074	16, 16	16, 16
FGA	20	1973	20, 22	21, 21
	21	2679
	22	1919
D8S1179	10	3237	10, 13	10, 12
	12	1297
	13	2116
D21S11	29	2077	29, 30	30, 30
	30	4266
D18S51	17	1637	17, 19	18, 19
	18	1101
	19	2468
D5S818	11	4825	11, 13	11, 12
	12	2154
	13	2600
D13S317	9	1774	11, 14	9, 12
	11	2173
	12	1566
	14	2016
D7S820	10	3528	10, 12	10, 10
	12	1559

[0240] The output of the web LSD application is shown in FIGS. 25A-C.

[0241] The LSD predicted genotype profile of each contributor suggested by the LSD method is shown in Table 33.

[0242]

TABLE 33

[0243] Among the nine loci, seven have 3 or 4 alleles per locus and the top-ranked genotype profiles for the two contributors correspond to the true genotypes.

[0244] At each of the remaining 2-allele loci, D21S11 and D7S820, two possible choices are selected because both choices in each locus give a mass ratio consistent with those for other loci. The ratio of the allele peak areas measured at each locus is close to 1:2 and the approximate mass ratio in this mixed DNA sample can also be close to 1:2 as predicted from the other loci. Thus two genotype combinations are possible with consistent mass ratios across both loci, and both fit equally well to the allele peak data measured, as shown in Table 33. Only one of these two is consistent with the true genotypes of the female victim and the male perpetrator. Therefore, when the victim's genotype's is brought in to resolve the choices, one would choose the one where person 2's profile agrees with that of the known victim's profile. Those are given as the shaded cells in Table 33.

[0245] The best-fit mass ratios across the nine loci are consistent with each other, ranging from 1:1.20 (d5S818) to 1:1.76 (D3S1358).

[0246] This set of mixture sample was successfully resolved at all loci. The victim's profile was used at two 2-allele loci to help select between two alternative genotype combination choices.

Example 8B

LSD Predicts the Genotype Profiles of Each Contributor in Texas DNA Mixture Data Number 4

[0247] The electropherogram showing the allele peak data at nine loci for Texas DNA mixture number 4 is shown in FIG. 26. The peak area data extracted from the electropherogram and the known genotype profiles of the female victim and male suspect are shown in Table 34.

[0248]

	TABLE 34

		Allele Peak
	Alleles in the	Areas	Contributors' Genotypes
Locus	Mixture	Measured	Female Victim	Male suspect

D3S1358	14	616	14, 18	D3S1358
	15	1463
	18	1886
VWA	11	862	15, 16	VWA
	15	186
	16	185
	19	788
FGA	20	312	20, 21	FGA
	21	282
	24	1064
	25	1215
D8S1179	10	821	13, 16	D8S1179
	13	227
	15	769
	16	188
D21S11	29	1080	30, 31.2	D21S11
	30	261
	31	949
	31.2	202
D18S51	12	1115	12, 14	D18S51
	14	962
D5S818	11	634	11, 13	D5S818
	12	328
	13	389
D13S317	8	1143	12, 12	D13S317
	12	558
	13	1027
D7S820	8	175	8, 9	D7S820
	9	205
	10	817
	11	679

[0249] The data were analyzed by the LSD method. The web-output of the data after LSD analysis is shown in FIGS. 27A-C. The LSD predicted genotype profile of each contributor suggested by the LSD method is shown in Table 35.

[0250]

TABLE 35

	Alleles	Peak Area	LSD Suggested Genotype Resolution Result
	in the	Ratio	Person 1	Person 2	Mass Ratio
Locus	Mixture	Measured	(Victim)	(Suspect)	Calculated	Remarks

D3S1385	14	1:	14, 18	15, 18	1:2.54	True one
	15	2.38:
	18	3.06
VWA	11	4.66:	15, 16	11, 19	1:4.45	True one
	15	1.01:
	16	1:
	19	4.26
FGA	20	1.11:	20, 21	24, 25	1:3.84	True one
	21	1:
	24	3.77:
	25	4.31
D8S1179	10	4.37:	13, 16	10, 15	1:3.83	True one
	13	1.21
	15
	16
D21S11	29	5.35:	30, 31.2	29, 31	1:4.38	True one
	30	1.29:
	31	4.70:
	31.2	1
D18S51	12	1.16:	12, 14	12, 14	Indeterminate	True one
	14	1
D5S818	11	4.20:	11, 13	11, 12	1:3.60	True one
	12	3.41:
	13	1
D13S317	8	2.05:	12, 12	8, 13	1:3.89	True one
	12	1:
	13	1.84
D7S820	8	1:	8, 9	10, 11	1:3.94	True one
	9	1.17:
	10	4.67:
	11	3.88

[0251] From the summary of the LSD-suggested genotype resolution result shown in Table 35, it is clear that all nine loci are resolved correctly and unambiguously, where the top-ranked genotype combinations selected by LSD are consistent with the true genotype profiles of the two contributors at all nine loci. The mass ratios calculated from the nine loci are also consistent, ranging from 1:2.54 (D3S1358) to 1:4.45 (VWA).

[0252] Deconvolution at locus D18S51 warrants further discussion. D18S51 has two alleles and, as always, three of the four possible genotype combination profiles have zero fitting errors. The best fitting mass ratios calculated from these three possible combinations are 1:1.2, 1:44.6, and 1:12.6, none of which are consistent with those calculated from the other eight loci. Therefore, none is selected as a possible choice for this locus. Since the ratio of the allele peak areas measured at this locus is 1.16:1, which is very close to 1:1, the genotype combination of {[12, 14][12, 14]} is considered as the possible choice. This combination was chosen in this instance, and it is the correct choice.

[0253] LSD is able to resolve DNA mixtures from the Acadiana Crime Lab and the Texas Department of Public Safety in Corpus Christi into the genotype profile of the two contributors. From these data it is apparent that as long as the allele peak data in a sample reflect the relative mass proportion of the alleles in the mixture, LSD will yield the true genotype profiles for the two contributors as its top-ranked profiles. Therefore, the LSD method is a feasible, powerful, and highly useful tool in resolving DNA mixture samples of two people.

Example 9

LSD is a More Accurate Method to Resolve a DNA Mixture into Genotype Profiles of Contributors than Linear Mixture Analysis (LMA)

[0254] Linear Mixture Analysis (LMA) is a recently published algorithm that attempts to resolve DNA mixtures. LMA, like LSD, uses quantitative allele peak data and linear algebra principals to solve the DNA mixture problem. LSD is a more flexible method for solving for the most fitting genotype profile for individual contributors in the mixture. LSD operates locus by locus to fit each locus separately, followed by pulling together only those loci at which resolution is clear and consistent to form a composite profile for each of the two contributors. In LMA, all available loci are processed as one entity, and a single mass ratio is sought to fit the given allele peak data simultaneously at all loci.

[0255] When LMA is used to resolve a two-people DNA mixture, the genotype of one of the two contributors also has to be known, and entered into the LMA algorithm to derive the other contributor's genotype. When using LSD to resolve such mixed DNA profiles, no a priori genotype information is necessary. The best-fit genotype combination pair for both contributors is obtained simultaneously in one step.

[0256] To verify that the LSD method is more accurate than the LMA method, data from a DNA mixture was resolved into genotype profiles using both methods. The results of each method were then compared to the known genotype profile of each contributor.

[0257] The data for the mixture of DNA used to test the accuracy of the LSD and LMA methods is shown in Table 36.

[0258]

TABLE 36

Alleles in		Genotype
the		Combination	Allele Peak
Locus	Mixture	Suspect	Victim	Areas

TH01	8	8, 9.3	9.3, 9.3	17441
	9.3			22368
D21S11	59	67, 70	59, 65	1226
	65			1434
	67			8816
	70			8894
D18S51	13	13, 13	16, 17	38985
	16			1914
	17			1991
D8S1179	10	10, 14	10, 11	6416
	11			383
	14			5659
VWA	16	16, 18	17, 19	4669
	17			931
	18			4724
	19			188
FGA	21	21, 22	21, 23	16099
	22			10538
	23			1014

[0259] Both the LMA and LSD methods were used to resolve the DNA mixtures into the genotype profile of each contributor. Table 37 shows the genotype profiles of the contributors resolved using the LSD method.

[0260]

TABLE 37

	Alleles	Peak Area	LSD Suggested Genotype Resolution Result
	in the	Ratio		Mass Ratio
Locus	Mixture	Measured	Person 1	Person 2	Calculated	Remarks

TH01	8	1:	9.3, 9.3	8, 9.3	1:7.08	With
	9.3	1.28				confidence
D21S11	59	1:	59, 65	67, 70	1:6.66	With
	65	1.17:				confidence
	67	7.19:
	70	7.25
D18S51	13	20.37:	16, 17	13, 13	1:9.98	With
	16	1:				confidence
	17	1.04
D8S1179	10	16.75:	10, 11	10, 14	1:11.39	With
	11	1:				confidence
	14	1.476
VWA	16	24.84:	17, 19	16, 18	1:8.39	With
	17	4.95:				confidence
	18	25.13:
	19	1
FGA	21	15.88:	21, 23	21, 22	1:4.76	With
	22	10.39:				confidence
	23	1

[0261] The LSD result shown in Table 37 correctly predicts the genotype profile for each contributor at all six loci. It is worth noting that the best-fit mass ratio calculated for the six loci varies from 1:4.76 (FGA) to 1:11.39 (D8S1179). This variation does not affect the ability of the LSD method to correctly predict the genotype profiles of the contributors.

[0262] Table 38 provides the genotype profiles predicted using the LMA method of analysis.

[0263]

TABLE 38

[0264] The LMA method only resolves two of the six loci correctly. The shaded boxes in the table indicate the incorrectly resolved loci.

[0265] A second set of DNA mixture sample data was used to compare the effectiveness of LMA and LSD in resolving 2-people mixtures. The data used to perform these analyses are shown in Table 39. The data were provided by the Promega Corporation and have a known mass ratio of 9:1 prior to PCR. The known genotypes of both contributors are also shown and were used to verify the resolution results.

[0266]

	TABLE 39

	Alleles in the	Ratio of Allele Peak	True Genotypes
Locus	Mixture	Data	Person 1	Person 2

TH01	6, 9, 9.3	19.79:1:1.19	6, 6	9, 9.3
D18S51	12, 13, 15, 17	14.23:1.40:11.34:1	12, 15	13, 17
D5S818	10, 11, 12	3.08:1:2.97	10, 12	11, 11
D13S17	9, 10, 11	1:5.65:5.10	10, 11	9, 10
D7S820	9, 11, 12	54.10:1.94:1	9, 9	11, 12
CSF1PO	10, 11, 12	1:11.74:12.42	11, 12	10, 12
VWA	15, 17, 18	3.89:1:4.28	15, 18	17, 17
D8S1179	12, 13, 15	4.40:1:3.99	12, 15	13, 13
FGA	21, 24, 25	10.58:12.76:1	21, 24	24, 25

[0267] The LSD suggested genotype resolution results are presented in Table 40.

[0268]

	TABLE 40

	LSD Suggested Genotype Resolution
	Result
				Mass Ratio
Locus	Alleles in the Mixture	Person 1	Person 2	Calculated

TH01	6, 9, 9.3	6, 6	9, 9.3	9.0:1
D18S51	12, 13, 15, 17	12, 15	13, 17	10.7:1
D5S818	10, 11, 12	10, 12	11, 11	6.0:1
D13S17	9, 10, 11	10, 11	9, 10	5.8:1
D7S820	9, 11, 12	9, 9	11, 12	18.4:1
CSF1PO	10, 11, 12	11, 12	10, 12	13.0:1
VWA	15, 17, 18	15, 18	17, 17	8.1:1
D8S1179	12, 13, 15	12, 15	13, 13	8.4:1
FGA	21, 24, 25	21, 24	24, 25	7.9:1

[0269] Table 40 shows that all nine loci are correctly resolved by pulling out the top-ranked genotype combination pair at each locus. Note that the optimal mass ratio calculated for each locus varies somewhat from locus to locus, ranging from 5.8:1 to 18:4:1, and the average being 9.7:1, which is very close to the known mass ratio prior to amplification.

[0270] In contrast, results from using the LMA algorithm, shown in Table 41, indicate that only seven out of nine loci were correctly resolved.

[0271]

TABLE 41

[0272] The LMA genotype resolution results at loci D18S51 and D7S820 are not correct. Note that these two loci correspond to those loci, from LSD, to have the best-fit mass ratio the most removed from the average ratio of 9.7:1, indicating the imposed fit used by the LMA approach for these loci is a poor fit, although it appear to be a good fit for other loci.

[0273] From these two examples, it appears that LMA is not as accurate as LSD in resolving mixed DNA samples.

[0274] While the invention has been described with respect to specific examples including presently preferred modes of carrying out the invention, those skilled in the art will appreciate that there are numerous variations and permutations of the above described systems and techniques that fall within the spirit and scope of the invention as set forth in the appended claims.

REFERENCES

[0275] 1 J. Samuels and C. Asplen, The Future of Forensic DNA Testing: Predictions of the Research and Development Working Group, The National Institute of Justice, November, 2000.
[0276] 2 National Research Council, The Evaluation of Forensic DNA Evidence, National Academy Press, 1996.
[0277] 3 T. Clayton, J. Whitaker, R. Sparkes, and P. Gill, “Analysis and Interpretation of Mixed Forensic Stains Using DNA STR Profiling,” Forensic Science International, 91 (1998): 55-70.
[0278] 4 J M. Curran, CM. Triggs, J. Buckleton, and BS. Weir, “Interpreting DNA Mixtures in Structured Populations,” Journal of Forensic Science, 44-5 (1999): 987-995.
[0279] 5 P. Gill, R. Sparkes, R. Pinchin, T. Clayton, J. Whitaker, and J. Buckleton, “Interpreting Simple STR Mixtures Using Allele Peak Areas,” Forensic Science International, 91 (1998): 41-53.
[0280] 6 T. Wang, N. Xue, M. Radar, and J. D. Birdwell, “Least Square Deconvolution (LSD) of Mixture STR/DNA Using Peak Area/Height Data,” The 7^thCODIS User's Conference, Arlington, Va., October, 2001.
[0281] 7 M. Farley and J. Harrington, Forensic DNA Technology, Lewis Publishers, INC., 1991.
[0282] 8 K. Inman and N. Rudin, An Introduction to Forensic DNA Analysis, CRC Press, 1997.
[0283] 9 P. Gill, J. Whitaker, C. Flaxman, N. Brown, and J. Buckleton, “An Investigation of the Rigor of Interpretation Rules for STRs Derived from Less Than 100 pg of DNA,” Forensic Science International, 112 (2000): 17-40.
[0284] 10 J. Butler, Forensic DNA Typing, Academic Press, 2001.
[0285] 11 B. Budowle, “Probabilities of Alleles' Distribution at 13 Loci”, Personal Communication in October 1997.
[0286] 12 P. Gill, R. sparkes, and C. Kimpton, “Development of Guidelines to Designate Alleles Using an STR Multiplex System,” Forensic Science International, 89 (1997): 185-197.
[0287] 13 B. Brown, Director of FBI National CODIS Project, Personal Communication in January, 2002.
[0288] 14 P. Billings, DNA on Trial, Genetic Identification and Criminal Justice, Cold Spring Harbor Laboratory Press, 1992.
[0289] 15 I. Evett, and B. Weir, Interpreting DNA Evidence, Statistical Genetics for Forensic Scientists, Sinauer Associates Inc., 1998.
[0290] 16 G. Strang, Linear Algebra and Its Applications (Third edition), Harcourt Brace Jovanovich College Publisher, 1986.
[0291] 17 J. Birdwell, “Analysis of Forensic Mixtures,” LIT/FBI Weekly Seminar Series to Project Team Members, Apr. 27, 2001.
[0292] 18 M. Perlin and B. Szabady, “Linear Mixture Analysis: A Mathematical Approach to Resolving Mixed DNA Samples,” Journal of Forensic Science, 46-6(2001): 1372-1378.
[0293] 19 P. Graham, “Allele Peak Data of DNA Mixture from Real Cases,” Texas Department of Public Safety, Corpus Christi, October, 2001.
[0294] 20 “Web-LSD Software: Restricted database, Developed by Laboratory for Information Technologies, the University of Tennessee, Knoxville, May, 2002.
[0295] 21 I. Evett, P, Gill, and J. Lambert, “Taking Account of Peak Areas When Interpreting Mixed DNA Profiles,” Journal of Forensic Science, 43-6 (1998): 62-69.
[0296] 22 R. Wickenheiser, “Allele Peak Data of DNA Mixture from Real Cases,” Acadiana Crime Laboratory, Louisiana, September, 2001.
[0297] 23 P. Newman, “Allele Peak Data of DNA Mixture”, Promega Corporation, September, 2001.
[0298] 24 Personal Communication with DNA analysts around the country and through our work in resolving mixed DNA samples.
[0299] 25. T. Wang, “Guidelines and Rules for Interpreting Least Square Deconvolution (LSD) Output Results”, University of Tennessee, Sep. 15, 2002.

Claims

What is claimed is:

1. A computer implemented method of remotely accessing a software application of a predetermined computer server specifically programmed to run the software application for resolving a mixture of DNA from more than one source responsive to a client device, the computer implemented method comprising:

hosting said software application on the specifically programmed computer server, the computer server for providing access to a genotype profile database;

accessing said software application from the client device remotely via a network;

transmitting allele peak data for alleles present at a first locus of a plurality of loci in the DNA mixture from the client device to the computer server; and

calculating an allele peak fit at the specifically programmed computer server for a possible genotype combination at a first locus and subsequent locus on a locus by locus basis to resolve the mixture of DNA from more than one source, the DNA mixture resolution for transmission to the client device.

2. The method of claim 1 wherein the computer server is a secure server.

3. The method of claim 1 further comprising receiving the DNA mixture resolution comprising a calculated peak fit ratio and fitting residual at the client device.

4. The method of claim 1 wherein the network is a public IP network.

5. The method of claim l wherein the network is a private wide area network.

6. The method of claim 1 wherein the allele peak data and mixture resolution is encrypted for transmission.

7. The method of claim 1 wherein the server uses an encrypted communications protocol.

8. The method of claim 1 wherein the allele peak data and mixture resolution are compressed for transmission.

9. The method of claim 1 that further comprises the computer server remotely accessing the genotype profile database.

10. The method of claim 9 wherein the genotype profile database is on a secure server.

11. The method of claim 9 wherein the genotype profile database is accessed via a network.

12. The method of claim 1 further comprising protecting the computer server via a firewall.

13. The method of claim 1 wherein the allele peak data for alleles present at a single locus in the mixture of DNA is derived from a sample comprising DNA from more than one source individual.

14. The method of claim 1 wherein the allele peak data comprise measurements of one of peak height and area.

15. The method of claim 3 wherein the fitting residual for a possible genotype combination.comprises a sum of squared differences between components of a quantitative allele peak data vector and components of a fitted allele peak data vector.

16. The method of claim 1 wherein the first locus is selected from the group consisting of CSF1PO, FGA, TH01, TPOX, VWA, D3S1358, D5S818, D7S820, D8S1179, D13S317, D16S539, D18S51 and D21S11.

17. The method of claim 13 wherein one source individual of said mixture is known and another source individual of said mixture is unknown.

18. The method of claim 1 comprising selecting a genotype combination with a small fitting residual for each locus.

19. The method of claim 1 at least one locus harbors a short tandem repeat.

20. The method of claim 1 wherein one locus comprises three alleles.

21. The method of claim 1 further comprising separately repeating transmitting and calculating for a second locus different from said first locus.

22. The method of claim 1 further comprising searching for a match to a resolved genotype combination for the mixture in the genotype profile DNA database.

23. The method of claim 3 further comprising separately computing said peak fit ratio and said fitting residual for each of a plurality of loci, locus by locus.

24. The method of claim 1 further comprising outputting a resolved genotype combination at said client device.

Patent Citations (15)

Patent	Date	Inventor	Cited By
US5161204(A)	1992-11-01	Hutcheson et al.	Applicant
US5273632(A)	1993-12-01	Stockham et al.	Applicant
US5374527(A)	1994-12-01	Grossman	Applicant
US5541067(A)	1996-07-01	Perlin	Applicant
US5580728(A)	1996-12-01	Perlin	Applicant
US5759369(A)	1998-06-01	Menchen et al.	Applicant
US5876933(A)	1999-03-01	Perlin	Applicant
US6054268(A)	2000-04-01	Perlin	Applicant
US6741983(B1)	2004-05-01	Birdwell et al.	Applicant
US6750011(B1)	2004-06-01	Perlin	Applicant
US6807490(B1)	2004-10-01	Perlin	Applicant
US2002/0086289(A1)	2002-07-01	Straus	Applicant
US2002/0152035(A1)	2002-10-01	Perlin	Applicant
US2003/0143554(A1)	2003-07-01	Berres et al.	Applicant
WO2003056494	2003-07-01		Applicant

Non-Patent Literature (10)

Perlin, Mark W. et al., “Linear Mixture Analysis: A Mathematical Approach to Resolving DNA Samples,” Journal Forensic Science, 2001, vol. 46, No. 6, pp. 1372-1378.Applicant
Wang, Tsewei, “Commentary,” Journal Forensic Science, Sep. 2002, published Aug. 21, 2002, vol. 47, No. 5, Paper ID JFS2002092—475.Applicant
Perlin, Mark W., “Author's Response,” Journal Forensic Science, Sep. 2002, published Aug. 21, 2002, vol. 47, No. 5, Paper ID JFS2002123—475.Applicant
Schwartz, Lisa S. et al., “Fluorescent Multiplex Linkage Analysis and Carrier Detection for Duchenne/Becker Muscular Distrophy,” Am. J. Human Genetics 51:721-729, 1992.Applicant
McConkey, E. H., “Human Genetics: The Molecular Revolution,” Jones and Bartlett Publishers, 1993, pp. 92-112.Applicant
Clayton, T. M. et al., “Analysis and Interpretation of Mixed Forensic Stains Using DNS STR Profiling,” Forensic Science International, vol. 91, pp. 55-70, 1998.Applicant
Gill P. et al., “Interpreting Simple STR Mixtures Using Allele Peak Areas,” 1998, Forensic Science International, vol. 91, pp. 41-53.Applicant
Evett, Ian W. et al., “Taking Account of Peak Areas When Interpeting Mixed DNA Profiles,” 1998, Journal of Forensic Sciences, vol. 43. No. 1, pp. 62-69.Applicant
Perlin, Mark W. et al., “Toward Fully Automated Genotyping: Genotyping Microsatellite Markers by Deconvolution,” 1995, American Journal of Human Genetics, vol. 57, pp. 1199-1210.Applicant
Wang, Tsewei et al., “Mixture STR/DNA Deconvolution Using Allele Peak Area Data and the Least Square Technique,” 12th International Symposium on Human Identification, Biloxi, MS, Oct. 9-12, 2001.Applicant