Machine Learning-Driven Analysis of Whole-Genome Data Reveals the Population Genetic Structure and Geographic Differentiation of Toxoplasma gondii

This study developed a high-resolution MLST method for Toxoplasma gondii genotyping using whole-genome SNP data and machine learning. XGBoost achieved an average accuracy of 97% in cross-validation, and feature selection identified a core set of informative SNPs for robust classification. Unsupervised clustering and PCA supported a three-genetic-cluster structure, which correlated strongly with geographic origin and aligned with internationally recognized typing: Cluster I (classical Type I/II), Cluster II (atypical strains), and Cluster III (Type III). This ML framework significantly enhances typing accuracy and resolution, providing a powerful tool for genomic epidemiology of T. gondii.