Validation of MaSTR™ Software (Mixture analysis from STRs): a fully-continuous probabilistic mixture analysis validation using PowerPlex®Fusion 2 – 5 contributor mixtures
Probabilistic genotyping (PG) for evaluation of mixtures provides many advantages compared to historical (CPI, binary likelihood ratio) approaches. Fully continuous
approaches utilize more of the information from the evidentiary profile; such as peak heights, stutter percentages, mixture ratio, probability of drop-in/out to provide
weighted genotypes from the mixture profile. The weighted genotypes are used to calculate a likelihood ratio that a profile is included, or not included, in the mixture.
Mixture profiles are compared to hypothetical profiles created based on these parameters and hypothetical combinations of genotypes present in the sample in order
to calculate how probable it is that a given individual contributed to the mixture. Integration over a large number of interrelated parameters is required to calculate
relative probabilities but doing so directly is not feasible. Marcov Chain Monte Carlo (MCMC) is a widely used sampling method to approximate such complex
integrals with high accuracy.1,3 Statistically sound, rigorous software tools are essential for forensic interpretation of complex mixtures that often have multiple
contributors, low-level DNA of one or more of the contributors, shared alleles, minor contributor alleles in stutter positions to major contributors and drop-out.
Thorough validation of methodologies is required for forensic laboratories. Conclusions of a preliminary evaluation of MaSTR software indicated MaSTR software
provides the rigor of a fully continuous probabilistic approach1 in a straight-forward software program. A full validation was proposed as the results of the
preliminary study were concordant with the known mixture contributors. The validation study of MaSTR software provides the forensic community with a detailed
report of the capabilities and reliability of this newly available tool to assist the forensic analyst in applying their expertise to evaluate mixture profiles.
The validation study was designed in accordance with the Scientific Working Group on DNA Analysis Methods (SWGDAM) and Organization of Scientific Area
Committees (OSAC) guidelines. Purified, de-identified DNA from 40 donors was obtained from the Nebraska Biobank (University of Nebraska Medical Center),
quantified using the Quantifiler Human DNA Quantification kit (Thermo Fisher), amplified with PowerPlex®Fusion (Promega Corporation) and analyzed on an ABI
PRISM® 3130 Genetic Analyzer (Thermo Fisher). Replicates of single source samples were genotyped (GeneMarker®HID software). Files were selected to
provide a range of allele calls and stutter information. Results of these replicates and their dilutions were imported into MaSTR software to calculate the variance
coefficient and stutter ratios required for the protocol data set. The protocol data set provides the software with context of expected peak height variation over the
range of potential RFU values and stutter during the evaluation of mixture profiles. Mixtures ranged from simple, two-contributors with major/minor components
with few shared alleles to complex four and five-contributor mixtures with a range of mixture ratios and shared alleles. Mixture samples were also diluted to
examine low-template effects across the range of contributors.
The study utilized the standard model within the software. This model utilizes peak heights, drop-in/out, degradation, stutter, co-ancestry (NRC I and NRC II) and allele
frequencies from USA populations.5 Tests included evaluation of mixtures with: 1) a known contributor profile and no person of interest (POI) profile(s), 2) a known
contributor profile and POI profile(s), 3) no known and no POI profiles(s) to evaluate deconvolution capabilities and 4) contributor and non-contributor samples to
evaluate exclusion capabilities. Additional parameters included use of the elimination database, containing some of the known contributor genotypes, to mimic
detection of staff profiles in a mixture in order to evaluate MaSTR’s capability to detect contamination. The number of MCMC iterations and chains required increased
with complexity of the mixtures. Results were consistent with known contributors and exclusions. Methodology and tabulated results will be presented.
Accuracy of inclusion and exclusion
Seventy-seven analyses of 2 – 5-person mixtures resulted in accurate inclusion and positive likelihood ratio for all tested samples. Fifteen of the analyses included non-contributor samples. All of the non-contributor samples were excluded, having negative likelihood ratio results.
Figure 4: A Ratio Plot showing the correlation between the ratio of a contributor to a mixture and the average ratio value reported by MaSTR when that contributor's profile was tested.
DNA Quantification, Mixture Construction, Amplification, Capillary Electrophoresis
All human DNA samples used in this study were purchased as de-identified extracts from the Nebraska BioBank, and were used according to the University of Nebraska-Lincoln’s Institutional Review Board’s (IRB) approved policies for ethical standards and methods for human testing. Sample collection and experimentation commenced only after the IRB review process was completed
DNA was extracted at the Nebraska BioBank from 40 donor samples using the Qiagen QIAcube and the QIAamp 96 DNA QIAcube HT kit optimized for downstream genomic assays. Addition of proK and AL buffer and incubation at 56°C resulted in lysis of the cells. The samples were added to the spin columns in the Qiacube for washing and elution of DNA. The DNA was washed via a salt wash (AW1) followed by a salt eliminating wash (AW2) then an ethanol wash. DNA eluted from the column via a low ionic elution buffer. The resulting DNA was quantified using the NanoDrop™ One microvolume UV-Vis spectrophotometer at the Nebraska BioBank
In order to further determine the concentration of DNA in each sample after they were delivered, the extracts were quantified using an Applied Biosystems 7500 real-time PCR system and the Applied Biosystems Quantifiler™ Human DNA Quantification Kit according to the manufacturer’s recommendations.
Prior to the analysis of mixed samples, a protocol data set was created for the MaSTR software to use in the assessment of stutter peaks and the calculation of a variance factor. Ten samples were chosen that demonstrated the greatest level of allelic variation available. Each sample was amplified using an initial quantity of 0.5 ng with five replicates. The samples were then diluted 1:2 and 1:4 and amplified again, each with five replicates, yielding a total of 150 analysis samples with 10 distinct genotypes and a broad range of peak heights for the MaSTR software to use.
Figure 5: A Likelihood ratio plot showing the highest LR reported for
contributor profiles that were tested as unknowns, showing that
contributors known ratio in the mixture, the total number of contributors,
and the amount of allele sharing amongst the contributors in the sample.
Mixture samples were created using two, three, four, or five of the single source samples. Two different strategies were used in creating the two and three-person
mixtures. After each single source sample was genotyped, an appropriate number of samples (two or three) were selected that showed minimal or maximal allele
sharing amongst the available genotypes, in order to test the MaSTR software under these different conditions. Two different two-person mixture combinations were
created, one in which the samples shared a total of five alleles amongst five loci (low share) and one in which the samples shared 19 alleles amongst 18 loci (high
share). The same process was used for the three-person mixtures, resulting in one combination with 16 alleles shared amongst 15 loci between the three contributors
(low share) and a different combination of contributors showing 31 alleles shared amongst 21 loci (high share). The four- and five- person mixtures were created by
randomly selecting previously unused samples and combining them without regard to allele sharing. This gave a total of 19 different single source genotypes used
to create 6 mixture genotypes, no genotype was used in more than one combination. Two-person mixtures were made in the following ratios; 1:1, 1:2, 1:3, 1:5, and
1:10. Three-person mixtures were made in the following ratios; 1:1:1, 1:1:2, 1:3:5, and 1:2:10. Four-person mixtures were made in the following ratios; 1:1:1:1,
1:2:2:5, and 1:1:3:10. Five-person mixtures were made in the following ratios; 1:1:1:1:1, 1:2:2:5:10, and 1:1:5:5:10. The mixture ratio balance for each sample was
assessed by multiple methods prior to final amplifications. Initially, the peak heights of each single source sample selected for combining were compared to the others
in the planned mixture. The amount of input DNA for each contributor was then adjusted so that the peak heights in the mixture should correspond to the desired
ratio. Test mixture samples were then assembled and run in triplicate to assess if the amount of DNA in the respective amplifications gave acceptable peak height
ratios or needed further adjustment. Once the test mixture samples demonstrated peak heights that fit expectations for the desired mixture ratio, those quantities of
DNA were used to create the experimental samples. Each mixture sample was also serially diluted 1:2, 1:4, and 1:8 in order to examine the same contributor ratios
with and without stochastic amplification effects. Every final mixture/dilution combination sample was run in triplicate.
Figure 6 & 7: After an analysis is completed in MaSTR software, the user clicks on the job name to open the result screen. As an example, we present the above example from a two-person 1:1 mixture without a known contributor specified. Select the name of the file of interest from the drop-down box. Sample 1653 (left) has a strong likelihood corresponding to the deconvoluted unknown 2. Sample 1678 has a strong LR corresponding to the deconvoluted unknown 1. Likelihood ratios are calculated from the weighted genotype results
All samples were amplified using the Promega PowerPlex® Fusion 5C kit following the manufacturer’s protocols. The target initial quantity of DNA for amplification
was 0.5 ng and was subsequently adjusted according to the requirements of the specific experimental conditions. The amplification reaction final volume was
25 µL for all samples and thermal cycling was performed in an Applied Biosystems GeneAmp PCR System 9700 using maximum ramp speed for 30 cycles. All
amplification sets included a positive control (cell line 2800M DNA, included in kit) and a negative control (deionized water used for sample dilutions).
Separation and detection of the PCR products was performed using an Applied Biosystems 3130 Genetic Analyzer with Foundation Data Collection software v3.0,
following all manufacture recommended protocols. All injections were performed at 3 kV for 5 seconds. Separation occurred with Applied Biosystems Hi-DiTM
formamide in Applied Biosystems POP4 Polymer. Analysis of all the injection products was performed using SoftGenetics® GeneMarker® HID v2.9.0 software. All of
the two-person mixture samples were analyzed with an analytical threshold of 50 rfu, as were the three-person mixtures except for those that were diluted 1:8, which
used an analytical threshold of 30 rfu. All of the four- and five person mixtures also used an analytical threshold of 30 rfu. The differing analytical thresholds reflected
the prevalence of low level peaks in each condition and the desire to include allele dropout in some of the combinations in order to further challenge the MaSTR
software at each mixture ratio.
Figure 1 & 2: An example of a standard model. The jobs submitted included the mixture sample text file and text files of single source genotypes to be
evaluated, without specifying any known single source genotypes. These settings were selected in order to mimic
the most difficult cases where a known contributor swab is not available. The analyses included simple 2-person
mixtures with very limited allele sharing, as well as complex 3 – 5-person mixtures. Two-person mixtures with up
to 70% allele sharing were used to mimic related individuals in a mixture. The NRC II 4.1 approach for co-ancestry
was selected and coefficient was 0.1 Fifteen mixture analyses included non-contributor single source samples to
evaluate exclusion capabilities.
Models were consistent with the exception of varying the number of MCMC iterations. Models with increasing
numbers of iterations were used to provide information on minimum and optimal number of MCMC iterations for
2, 3, 4 and 5 contributor mixtures.
Figure 8: Weighted genotype results are viewed by selecting the Genotype Set Results from Results
tool bar. Each locus is displayed by selecting from the drop-down list. A heat map indicates the most
likely genotype combinations of the tested samples on a scale of 0 – 1; least likely (dark purple) to most
likely (yellow). The weighted genotype combinations are displayed in a table and are used in calculating
Experimental data have shown that the MaSTR software gives accurate Likelihood Ratio results when used with 2-, 3-, and 4-person mixtures. The calculated values
for each contributor LR fit well with the true identities and mixture ratios of the synthesized samples. All true-exclusion known samples gave negative (un-supporting)
LRs for the appropriate unknown “position” in MaSTR.
The “Major” and “Minor” contributors (mixture ratios of at least 3:1) were clearly and accurately distinguishable via their calculated LRs. Mixture contributors that were
at or close to a 1:1 ratio were accurately distinguishable in 2-person mixtures with no identified known sample in mixtures with a high degree of allele overlap. In 3-
and 4-person mixtures with a high degree of allele overlap, contributors present in near equal amounts with no identified known samples showed LRs accurately supporting
both samples being part of the mixture, but were sometimes difficult to tell apart
Preliminary work shows that using 10,000 MCMC iterations with eight chains works well for 2-person mixtures. 3-person mixtures, especially with minor contributors,
worked well using 20,000 MCMC iterations with eight chains and 4+ person mixtures showed the most consistent results running 50,000 MCMC iterations with eight
chains. Increasing the number of MCMC iterations gave more consistent LR values across the replicate MaSTR runs, especially with samples containing minor contributors,
allele dropout, shared alleles, and close ratio contributor data
MaSTR software runs using less than 100K total MCMC iterations were resolved within 15 minutes or less, while longer runs (400K) were completed within 45 minutes
using a desktop computer (Intel® Core™ i7-7700 cpu 3.60 GHz 4 core 8 logical processors) with 8 GB RAM operating with Windows 10.
Further interrogation of the data set is planned to evaluate likelihood ratio results and analysis time when known contributor(s) are included in the analysis procedure,
as well as peak height variation in replicate samples.
We thank Promega Human Identity Group for providing the PowerPlex Fusion and matrix kits
- SWGDAM Guidelines for Validation of Probabilistic Genotyping Systems, 2015 and 2016 NIST draft on guidelines for validating probabilistic genotyping software.
- Organization of Scientific Area Committees (OSAC) 2016 Validation Standards for Probabilistic Genotyping Systems Draft
- Taylor D., Bright J.-A., Buckleton J.S. 2013. The interpretation of single source and mixed DNA profiles, Forensic Sci. Int.:Genetics 7:516-528
- Gill P., Gusmão L., Haned H., Mayr W.R., Morling N., Parson W., et al., (2012) DNA commission of the International Society of Forensic Genetics: Recommendations on the evaluation of STR typing results that may include drop-out and/or drop-in using probabilistic methods, Forensic Sci. Int:Genetics 6:679-688
- Hill C.R., Duewer, D.L., Kline, M.C., Coble, M.D., Butler, J.M. Forensic Science International: Genetics 7 (2013)