Application of machine learning for ancestry inference using multi-InDel markers

Ancestry inference through population stratification plays an important role in forensic applications. Specifically, ancestry information inferred from forensic DNA evidence can provide vital clues for criminal investigations. Current advances in ancestry inference mostly focus on ancestry informative markers. Hereinto, multi-InDel was proposed as one of the compound markers performing well in complex ancestral classification in the subpopulation of Asia. However, research on analytical methods necessary to make reliable predictions is lacking. The newly proposed compound markers could be assessed with alternative methods. In this study, promising discriminant methods were explored using multi-InDel markers for forensic ancestry inference. As a prerequisite, the adopted multi-InDel markers were assessed by classical methods for population genetics, such as FST analysis, MDS and STRUCTURE. In addition, dimensionality reduction methods and serial reduction strategies were applied for data visualization. Subsequently, machine learning methods, including logistic regression (LR), support vector machine (SVM), k-nearest neighbors (KNN) and extreme gradient boosting (XGBoost), were evaluated by diverse approaches. As the result of multifarious analyses through comparisons and estimations, XGBoost with one-hot encoding was shown to be more effective in population stratification and ancestry inference for challenging cases with admixed populations.