Inference genetic structure and regional ancestry of Asian populations using 164 AISNPs and machine learning approaches.

Ancestry-informative SNPs are widely used for biogeographical ancestry inference, yet relatively few studies have explored strategies to maximize ancestry inference potential of existing AISNP panels. Here, we evaluated patterns of genetic differentiation among Asian populations and the utility of a single panel of 164 AISNPs for group-level regional ancestry inference using machine learning (ML) approaches. Genotype data from 2285 individuals representing 16 Asian populations were analyzed using conventional population genetic methods and six supervised ML algorithms. Population genetic analyses consistently revealed two major clusters corresponding to Southwest Asia and East Asia (EA), together with a north-south genetic cline within EA. The Myanmar population exhibited a distinct genetic profile characterized by broad dispersion in principal component analysis and an additional ancestry component in STRUCTURE analysis. Among the evaluated classifiers, LightGBM achieved the highest performance for both three-group classification (macro F1-score = 0.903 ± 0.007) and East Asian two-group classification (macro F1-score = 0.892 ± 0.008). Several markers, including rs3811801, rs3827760, rs1426654, rs260690, and rs174570, showed concordance between traditional population genetic statistics and ML-based feature importance measures. These findings demonstrated that the utility of the 164 AISNP panel for capturing broad-scale genetic patterns and regional genetic differentiation among Asian populations. They also highlight that the informativeness of AISNPs varies according to the hierarchical level of population differentiation, providing insights for the optimization of forensic ancestry inference panels.