Fine-scale biogeographical ancestry inference in Southeast and East Asians via high-efficiency markers and machine learning approaches

Biogeographical ancestry inference offers valuable clues for forensic cold cases, but limited information is typically obtained from substructured populations within continental East Asian and Southeast groups. This study presents an integrative genomic dataset of 3,461 individuals from East Asia and Southeast Asia to elucidate the fine-scale population substructure and its role in precision forensic medicine. Six nested panels were developed with increasing ancestry-informative marker (AIM) density (ranging from 50 to 2,000 SNPs) to distinguish fine genetic differences between the six language groups and populations within the Sino-Tibetan language family. We found that the 2000 AIM panel exhibited differentiation efficiency in PCA comparable to that of all loci. Additionally, we constructed a classification machine learning model with an average prediction accuracy of 84%, highlighting the critical role of geographical information in improving model accuracy. Furthermore, we validated the accuracy of the deep learning method Locator in predicting geographical coordinates solely based on genetic information. This work highlights the power of integrating genetic and geographic data with artificial intelligence to refine fine-scale biogeographical ancestry inference, offering more profound insights into population structure in East Asia and Southeast Asia, with significant implications for forensic applications.