Probabilistic Genotyping and AI in Crime Labs

A DNA mixture may contain genetic material from multiple people. Traditional interpretation can become difficult when contributors are low-level, the profile is incomplete, or alleles are masked by overlap. Probabilistic genotyping software uses biological models and statistical calculations to evaluate how likely the observed data are under competing propositions. The output commonly includes a likelihood ratio (LR), comparing the probability of the evidence if one proposition is true with its probability under an alternative. NIST’s 2024 Scientific Foundation Review explains DNA mixture interpretation, likelihood-ratio frameworks, probabilistic genotyping, reliability, relevance, and validation. It emphasizes that mixtures create questions about contributor number, contamination, transfer, and whether detected DNA is relevant to an activity. NIST’s explainer notes that software, configuration, and model choices can affect results and that different laboratories may reach different outcomes on the same sample. Some mixtures may be too complex to interpret reliably. An LR is not the probability that a suspect contributed DNA, much less the probability that the suspect committed a crime. It is conditional on propositions and model assumptions. A source-level question might compare “the person contributed” with “an unknown unrelated person contributed”; it does not establish when or how DNA was deposited. Activity-level questions require additional reasoning about transfer, persistence, and circumstances. Expert testimony should make those levels clear. Validation is essential. A laboratory should test the software and its own procedures on mixtures representative of the casework it accepts, including relevant contributor counts, DNA quantities, degradation, and known ground truth. Analysts need training in the statistical and biological model, access to documentation, and controls for software version and configuration. Reports should identify propositions, assumptions, limitations, and interpretation boundaries. Independent review and disclosure of relevant validation information allow courts and opposing experts to examine reliability. Probabilistic genotyping can provide a structured evaluation of complex data, but the number does not replace scientific judgment or decide guilt.