Revisiting adaptive introgression at the HLA genes in Lithuanian genomes with machine learning.

We are writing to discuss the article titled ‘Disentangling Archaic Introgression and Genomic Signatures of Selection at Human Immunity Genes,’ published by this journal (Urnikyte et al., 2023). This study employed an ad-hoc approach, first applying the machine learning tool, ArchIE (Durvasula and Sankararaman, 2019), to detect introgression candidates, followed by the use of the iHS statistic (Voight et al., 2006) to identify candidates under positive selection. According to this study, the HLA-C gene displays both introgression and positive selection signals, suggesting it as a candidate for adaptive introgression in Lithuanians. However, this approach is problematic due to the varying effectiveness of the methods employed (Zhang et al., 2023) and the confounding effects of introgression on methods used to detect selection (Racimo et al., 2015). More specifically, adaptive introgression can be confounded by long-term balancing selection (Fijarczyk and Babik, 2015), and the human leukocyte antigen (HLA) genes, which encode the major histocompatibility complex (MHC) in human, are well known examples for long-term balancing selection (Andrés et al., 2009; Gelabert et al., 2024). Considering this, we reanalyzed the Lithuanian genomic data using a recently developed machine learning approach, MaLAdapt (Zhang et al., 2023), which is specifically designed to detect adaptive introgression through supervised learning. Our results suggest that the HLA-C gene is not a candidate for adaptive introgression.