SMART-MHmix: A probabilistic model for microhaplotype-based forensic DNA mixture analysis

The interpretation of complex DNA mixtures remains a persistent challenge in forensic genetics. Although probabilistic genotyping systems for capillary electrophoresis (CE)-based short tandem repeats (STRs) represent significant progress, their applicability is constrained by the inherent limitations of CE platform when analyzing highly complex mixtures. Microhaplotypes (MHs) are multi-allelic markers compatible with next-generation sequencing (NGS) and present a promising alternative. However, the absence of methods for quantitatively interpreting NGS-based MH data hinders their practical application. To address this gap, we present SMART-MHmix, a probabilistic genotyping framework designed to model mixture profiles from NGS-sequenced MH loci and perform statistical evidentiary assessment. The framework calculates the probability of observed allele read counts conditional on their expected values, where the expected read counts are determined from the product of contributor template DNA amounts, locus-specific amplification efficiencies, and replicate-specific effects. The core likelihood calculation integrates three key components: a log-normal distribution modeling true allele signals, probabilistic models for stochastic drop-in and drop-out events, and Markov Chain Monte Carlo (MCMC) sampling for robust inference. We evaluated the performance of SMART-MHmix using 104 synthetic DNA mixtures of 2 to 5 contributors, profiled with the MHSeqTyper47 kit (47 autosomal MH loci). SMART-MHmix demonstrated robust performance across all mixture complexities. Likelihood ratio (LR) analysis provided strong discriminating power, with LRs for true contributors reaching up to 10⁴⁷ and over 86% exceeding 10⁵ across all mixture complexities, and correct support for exclusion for non-contributors. Using a threshold of θ=1, both sensitivity and specificity exceeded 90% in all tested scenarios. Mixture deconvolution enabled accurate genotype inference for major contributors, with high values for both the number of resolved loci and match success. Inference of the second contributor was reliable when the major contributor proportion exceeded 60%. For minor contributors ranked third or lower (by template DNA proportion), both the number of resolved loci and match success declined as mixture complexity increased. Overall, SMART-MHmix represents a dedicated continuous probabilistic model for MH-NGS data, enabling reliable analysis of complex forensic DNA mixtures with the potential to outperform conventional CE-STR workflows.