Multi-scale haplotype signatures distinguish among evolutionary processes

Abstract

Natural selection embeds distinct genomic signatures as populations adapt to environmental pressures. Machine learning approaches, particularly deep neural networks, have emerged as powerful tools to detect adaptive signatures from genomic data. While machine learning has dramatically improved our ability to extract these signals from sequence variation, distinguishing among multiple evolutionary processes remains challenging. We introduce SHADES, a deep learning framework that employs information about sequence differences and their frequencies at various spatial resolutions to capture evolutionary signatures at different genomic scales. We found that SHADES has the capacity to distinguish among neutrality, selective sweeps, balancing selection, and adaptive introgression and outperformed several contemporary methods evaluated for both classification accuracy and power. In addition, SHADES remained robust to a number of confounding factors, with only modest changes in classification performance. Furthermore, application of SHADES to empirical haplotypic variation from European humans recovered previously reported signals for selective sweeps, balancing selection, and adaptive introgression, while also identifying novel genes broadly involved in immune defense, inflammatory regulation, and epithelial barrier biology.