Probabilistic genotyping is a computerized method of interpreting DNA evidence that uses statistical modeling to make sense of samples containing genetic material from two or more people. It replaces the subjective judgment calls human analysts used to make about complex DNA mixtures with a mathematical process that produces a likelihood ratio: a number expressing how much more probable the observed DNA data is if a particular person contributed to the mixture than if an unknown, unrelated person did. The technology fills a real gap in forensic science, but its results depend on analyst decisions, validation boundaries, and sample conditions that are not always visible to the jury hearing the number.
What the Software Is Actually Doing
A DNA mixture is a sample containing genetic material from two or more people.1National Institute of Standards and Technology. DNA Mixtures: A Forensic Science Explainer These samples show up constantly in criminal cases. Swab a steering wheel, a doorknob, a weapon, or a piece of clothing, and the resulting sample will often contain DNA from several people who touched the object at different times. Traditional DNA analysis worked well for clean single-source samples. It struggled badly with mixtures.
When a lab amplifies a DNA sample, the result is an electropherogram showing peaks at various locations across the genome. For a single-source sample, the pattern is straightforward. For a mixture, the peaks overlap, stack, and blur together. A human analyst looking at a four-person mixture faces thousands of possible combinations of who contributed which peaks. Before probabilistic genotyping, analysts either made subjective calls about those peaks or declared the sample inconclusive.2National Institute of Justice. When DNA Samples Are Complicated: Calculating Variation in Mixed Samples Interpretation
Probabilistic genotyping software replaces that human analysis with a mathematical process. The two most widely used programs in the United States are STRmix and TrueAllele.3Federal Judicial Center. Probabilistic Genotyping Systems for Low-Quality and Mixture Forensic Samples Both are fully continuous models, meaning they use the actual height of every peak in the electropherogram rather than simply noting whether a peak is present or absent.4PubMed Central. A Review of Probabilistic Genotyping Systems: EuroForMix, DNAStatistX and STRmix
Both programs rely on a sampling technique called Markov Chain Monte Carlo. In plain terms, the software proposes a random combination of possible contributor genotypes, checks how well that combination explains the observed data, then adjusts and proposes another combination. It repeats this millions of times, gradually zeroing in on which genotype combinations are most consistent with the peaks in the electropherogram. The result is not a single answer but a probability distribution across many possible explanations for the mixture.
The software is not fully automatic. Before it runs, a human analyst must set several parameters, and biological artifacts in the data complicate the picture. Stutter peaks appear as smaller echoes of true peaks, caused by the copying enzyme slipping during replication. Allele dropout occurs when a contributor’s DNA is present in such a small quantity that one or both copies of an allele fail to amplify. The software uses biological models to account for these distortions, but misclassifying a true allele as stutter, or vice versa, changes which genotype combinations the software treats as plausible.
How to Read the Likelihood Ratio
Probabilistic genotyping software does not declare that a suspect’s DNA “matches” a sample. It produces a likelihood ratio comparing two competing explanations. The first, typically proposed by the prosecution, is that the suspect contributed DNA to the mixture. The second, aligned with the defense, is that an unknown, unrelated person contributed instead. The ratio tells you how many times more probable the observed data is under the first explanation than the second.5National Institute of Justice. Population Genetics and Statistics for Forensic Analysts – Likelihood Ratio
A likelihood ratio of 1 means the data is equally consistent with both explanations. A ratio of 1,000 means the data is a thousand times more probable if the suspect contributed than if a random person did. A ratio of 1,000,000 pushes into territory most scientists consider very strong evidence of contribution. A high number does not prove the suspect committed the crime; it quantifies the weight of the DNA evidence alone.5National Institute of Justice. Population Genetics and Statistics for Forensic Analysts – Likelihood Ratio
Because raw numbers can mislead jurors, the Department of Justice publishes a standardized verbal scale that forensic examiners may use alongside the numerical value:6Department of Justice. Uniform Language for Testimony and Reports for Forensic Autosomal DNA Examinations Using Probabilistic Genotyping Systems
- 1: Uninformative
- 2 to less than 100: Limited support
- 100 to less than 10,000: Moderate support
- 10,000 to less than 1,000,000: Strong support
- 1,000,000 or greater: Very strong support
When an examiner uses these labels, the DOJ requires the full scale to appear in the laboratory report so the jury can see where the number falls in context. The verbal description must always accompany the actual number, never replace it. Likelihood ratios close to zero can give an undue sense of certainty to people unfamiliar with the scale, because a small positive number feels definitive to someone who does not realize 1.0 is the neutral baseline.3Federal Judicial Center. Probabilistic Genotyping Systems for Low-Quality and Mixture Forensic Samples
Where the Results Can Go Wrong
Probabilistic genotyping is a real improvement over subjective human interpretation. It is not infallible, and several known issues affect how much confidence belongs with any given result.
The Contributor Count
The most consequential parameter the analyst sets is the estimated number of contributors to the mixture.3Federal Judicial Center. Probabilistic Genotyping Systems for Low-Quality and Mixture Forensic Samples If the analyst estimates three contributors but the sample actually came from four, every downstream calculation sits on a wrong assumption. Determining the true number of contributors from a messy electropherogram is genuinely difficult, and the software cannot figure it out on its own.
Research on real casework samples has shown that getting the contributor count wrong can shift the resulting likelihood ratio by a factor of 10,000 or more.7PubMed Central. The Impact of Considering Different Numbers of Contributors in Identification Problems Involving Real Casework Mixture Samples Underestimating the number tends to cause larger errors than overestimating it. The software also assumes by default that contributors are unrelated to each other. If the actual contributors are family members who share more genetic similarity than strangers, the results can skew in ways the software does not flag.3Federal Judicial Center. Probabilistic Genotyping Systems for Low-Quality and Mixture Forensic Samples
Validation Boundaries
A 2016 report from the President’s Council of Advisors on Science and Technology found that the foundational validity of probabilistic genotyping had been established only under narrow conditions: three-person mixtures where the minor contributor made up at least 20 percent of the total DNA and where the overall DNA quantity exceeded the method’s minimum threshold.8The White House. Forensic Science in Criminal Courts: Ensuring Scientific Validity of Feature-Comparison Methods For mixtures with more contributors, lower DNA quantities, or a minor contributor making up less than 20 percent of the sample, the report concluded that substantially more evidence was needed. The validated range has likely expanded since 2016, but the PCAST finding remains a benchmark for what has been rigorously tested versus what is extrapolated.
Each laboratory is also supposed to use the software only on mixture types for which it has conducted its own internal validation studies, under FBI quality assurance standards and standards from the Organization of Scientific Area Committees for Forensic Science.9Federal Bureau of Investigation. Quality Assurance Standards for Forensic DNA Testing Laboratories10National Institute of Standards and Technology. Two New Forensic DNA Standards Added to the OSAC Registry A lab that validated its process for three-contributor mixtures should not be analyzing casework from four or more contributors. Whether that boundary is actually honored at the bench is harder to verify from the outside.
Software Disagreement and Reproducibility
Different probabilistic genotyping programs use different mathematical models, and they do not always agree. Analyzing the same sample with different software can yield contradictory results, including cases where one program includes a person as a likely contributor and another excludes them.3Federal Judicial Center. Probabilistic Genotyping Systems for Low-Quality and Mixture Forensic Samples Even running the same sample through the same software twice will not always produce an identical likelihood ratio, because the Markov Chain Monte Carlo process involves random sampling that introduces slight variation with each run.
A 2024 NIST study examined how three laboratories using the same software handled 265 identical DNA mixture files. Over 92 percent of the resulting likelihood ratios fell within one order of magnitude of each other. For certain samples the variation was far larger, and the researchers identified five distinct causes of poor precision that arose despite identical inputs.11National Institute of Standards and Technology. DNA Mixture Interpretation: A NIST Scientific Foundation Review When different detection platforms were used on the same samples, nearly a third showed likelihood ratio differences of ten or more orders of magnitude.
Touch DNA and Secondary Transfer
Much of the evidence processed by probabilistic genotyping comes from touch DNA, the trace amounts of genetic material left behind when someone handles an object. Touch samples are especially prone to the problems that stress these systems: low quantities, degradation, and contributions from multiple unknown people. Your DNA can also end up on an object you never touched if someone who recently contacted you then handles that object. Paramedics, for example, can inadvertently carry DNA from one patient to another. Extraneous DNA shows up on clothing from laundering, on shared surfaces from routine daily contact, and on objects from prior users.12PubMed Central. Touch DNA Sampling Methods: Efficacy Evaluation and Systematic Review
Probabilistic genotyping can tell you the statistical weight of a DNA association. It cannot tell you how or when the DNA was deposited. A high likelihood ratio placing someone’s DNA on a weapon does not distinguish between the person who wielded it and someone whose skin cells transferred there through an intermediary hours earlier.
How Courts Decide Whether to Admit It
Before probabilistic genotyping results reach a jury, a judge must decide whether the evidence is legally admissible. Most federal courts and a majority of states apply the Daubert standard, under which the trial judge evaluates whether the method is testable, whether it has been peer-reviewed, whether it has a known error rate, and whether it is generally accepted in the relevant scientific community.13National Institute of Justice. Daubert and Kumho Decisions A smaller number of states follow the older Frye standard, which asks only whether the method has gained general acceptance in the relevant scientific community.
Probabilistic genotyping has been admitted under both standards in many jurisdictions. Rulings are not uniform. In People v. Wakefield (2015), a New York court found that TrueAllele’s probabilistic methods were generally accepted and superior to older approaches. For a different program called the Forensic Statistical Tool, judges in the same state split: People v. Collins (2015) ruled it inadmissible while companion cases admitted it.
People v. Hillary illustrates a deeper problem. The same DNA evidence was analyzed by both STRmix and TrueAllele, and the two programs produced conflicting results, one including the defendant as a contributor and the other not. The trial court excluded the probabilistic genotyping evidence, though the ruling turned on the laboratory’s lack of sufficient internal validation studies rather than the software disagreement itself. Admissibility often depends less on the science in the abstract than on whether the specific laboratory followed proper validation procedures for the specific type of mixture at issue.
The Source Code Fight
The most contentious legal battle surrounding probabilistic genotyping is whether defendants can inspect the software’s source code. Defense attorneys argue that the Sixth Amendment’s right to confront the evidence used against you requires access to the actual computer instructions that produced the likelihood ratio. A bug, an incorrect assumption, or a coding error that skews results can only be discovered by reading the code.
Software developers have resisted these requests, claiming the code is a trade secret. Courts have split on the issue. In United States v. Ellis and State v. Pickett, courts ruled that defendants were entitled to TrueAllele’s source code. In other cases, judges have denied disclosure, finding that cross-examining the software’s developer provides a sufficient substitute. When courts do order disclosure, they typically use protective orders that restrict access to defense experts who agree not to share the code publicly.3Federal Judicial Center. Probabilistic Genotyping Systems for Low-Quality and Mixture Forensic Samples The outcome often depends on an individual judge’s view of how far confrontation rights extend into algorithmic territory.
How This Evidence Gets Challenged
Defense attorneys have several avenues beyond source code access. Understanding them matters whether you are a defendant, a juror, or simply trying to evaluate a case in the news.
- Contributor count. If the analyst estimated the wrong number of contributors, the entire analysis is compromised. A defense expert can argue the electropherogram supports a different count and show how the likelihood ratio changes dramatically under that alternative.7PubMed Central. The Impact of Considering Different Numbers of Contributors in Identification Problems Involving Real Casework Mixture Samples
- Validation scope. If a lab validated its protocol for three-person mixtures but analyzed a five-person mixture in the defendant’s case, the results exceed the tested boundaries.10National Institute of Standards and Technology. Two New Forensic DNA Standards Added to the OSAC Registry
- Related contributors. The software’s default assumption that all contributors are genetically unrelated fails in cases involving family members. If the prosecution’s theory involves related individuals, the defense can challenge whether the analysis accounted for the relatedness.3Federal Judicial Center. Probabilistic Genotyping Systems for Low-Quality and Mixture Forensic Samples
- Transfer rather than presence. A high likelihood ratio only establishes a statistical association between a person’s DNA and a sample. It says nothing about when or how the DNA got there. In touch DNA cases, secondary transfer is a legitimate alternative explanation that the software cannot evaluate.
- Reproducibility. Running the same data through the same software twice may produce different likelihood ratios due to the randomness inherent in Monte Carlo sampling. A defense expert can request multiple runs and highlight the variation to argue the result lacks precision.3Federal Judicial Center. Probabilistic Genotyping Systems for Low-Quality and Mixture Forensic Samples
The strongest challenges tend to combine several of these points. Showing that the analyst guessed wrong on the contributor count and that the lab never validated the software for that mixture type is far more persuasive than either argument alone.