A biogeographical ancestry inference pipeline using PCA-XGBoost model and its application in Asian populations

Abstract

Biogeographical ancestry (BGA) inference plays a crucial role in genetics, anthropology, forensic science, and medical research. Current methods like principal component analysis (PCA) and ADMIXTURE, based on single nucleotide polymorphisms, are commonly used. Here, we introduce a bio-geographical ancestry inference pipeline that integrates prior population structure and clustering. Our pipeline first analyzes genetic structure on cleaned data to obtain optimal parameters and classification model labels. An XGBoost (eXtreme Gradient Boosting) classification model is constructed using principal components from PCA, and model predictions are evaluated with LR (likelihood ratio). The pipeline was applied to a dataset of Asian populations, with a first prediction accuracy of 96.27 % achieved. The LR-based evaluation accuracy reached 98.96 %, showing an improvement of 2.69 % with the introduction of LR assessment. This highlights the robust predictive capability of our pipeline and the improved accuracy in evaluation with LR. This successful application will benefit genetic research, human history studies, and criminal investigations. Additionally, the pipeline's versatility allows application to new datasets.