Abstract
BACKGROUND: Recent advances in population-specific high-quality reference databases have significantly improved the performance of forensic panel development for personal identification, parentage testing, and biogeographical ancestry inference. However, the discriminative power of previously developed AISNP panels remains limited in applications involving regional Chinese population substructures. RESULTS: We used the high-quality Chinese population-specific genetic resource to develop six nested C5ClusterTag-50/100/250/500/1000/2000 ancestry-informative SNP panels focused on inferring population stratification among geographically and genetically distinct Chinese populations. We used comprehensive bioinformatics approaches and machine learning to validate the effectiveness of these panels in both the testing and training datasets. A total of 2,772 individuals were screened across different AISNP panels based on the I(n) value. Ancestry inference power was assessed via principal component analysis (PCA), ADMIXTURE, and uniform manifold approximation and projection (UMAP) across the 2000-AISNP, 1000-AISNP, 500-AISNP, 250-AISNP, 100-AISNP, and 50-AISNP panels. The 1000-AISNP panel demonstrated the optimal balance between the SNP count and discriminative power for forensic applications. The accuracy of the random forest model was confirmed through a confusion matrix based on machine learning. CONCLUSIONS: These panels can differentiate ethnolinguistic Chinese populations into five subgroups based on geographical divisions or linguistic affiliations, achieving a high average accuracy rate in machine learning models. This work not only developed a robust ancestry inference panel and new tools for predicting the ancestry of ethnolinguistic Chinese populations but also created a comprehensive reference dataset and machine learning model applicable to population and forensic uses globally.