A Machine Learning Approach for Detecting Selective Sweeps Using Ancient DNA

Abstract

Biological adaptation leads to speci c patterns in population genetic data called selective sweeps. Although researchers have applied machine learning to sweep detection, which speci c methods are appropriate for any given scenario is not well understood. We conducted a systematic review of a suite of machine learning(ML) classi ers for sweep detection. We found that accurate models can be built using simple, fast classi ers supported by preprocessing. We produced a ML work ow which is applicable for general population genetic problems. Our methods were extended for ancient DNA, showing a sweep signal can be retrieved even at high missing rates.