Identifying unknown contributors to two-person DNA mixtures via whole-genome sequencing

Forensic investigative genetic genealogy (FIGG) has demonstrated considerable promise in identifying perpetrators and unidentified human remains. While whole-genome sequencing (WGS) offers a robust foundation for FIGG, DNA mixtures-prevalent in forensic casework-remain a challenge as standard algorithms require single-source profiles. Potential solutions fall broadly into two categories: (1) deconvoluting mixtures into genotype calls (which can be susceptible to allelic drop-out and miscalls in imbalanced scenarios), and (2) analyzing mixture profiles directly using likelihood ratio (LR) frameworks (which, though currently incompatible with genealogical database search algorithms, are powerful for kinship verification). Utilizing the DNBSEQ platform, we performed WGS on two-person mixtures across varying ratios (1:1, 1:2, 1:5, and 1:9), generating genome-wide data comprising an average of 4.34 million SNPs per sample. Our findings yield several practice-oriented insights: First, the hidden Markov model (HMM)-based Allele-Haplotype Hybrid Assignment (AH-HA) algorithm outperforms the locus-based EuroForMix (EFM) in deconvolution-especially for minor contributors-by effectively leveraging haplotype information, conditioned on one known contributor. Second, for major contributors, deconvolution error rates remain below 10%, enabling the identification of up to 6th- to 7th-degree relatives using error-aware algorithms (e.g., clusIBD) without imputation. Third, for minor contributors, despite error rates exceeding 10%, integrating imputation with the KING algorithm and empirical kinship correction facilitates the resolution of 3rd- to 4th-degree relatives. Finally, EFM is validated as a reliable confirmatory tool for direct kinship verification undeconvoluted profiles. Collectively, this work provides a comprehensive methodological comparison and proposes practical solutions to advance genetic genealogy analysis based on DNA mixture data.