Evolutionary and Population-Aware AI Models for Characterizing Global Diversity of Polymorphic Immune Genes Across Human Populations

Abstract

Polymorphic immune genes, particularly the human leukocyte antigen system, exhibit extraordinary diversity across global human populations, shaped by millions of years of evolutionary pressure from pathogens, environmental factors, and demographic history. This diversity underpins critical differences in disease susceptibility, vaccine response, and transplantation compatibility, yet existing computational approaches for characterizing these genes are largely built on datasets dominated by individuals of European ancestry and fail to incorporate population structure or evolutionary dynamics. This paper presents a system-level examination of evolutionary and population-aware artificial intelligence models designed to characterize global diversity of polymorphic immune genes across human populations. We propose that such models must integrate principles from population genetics, evolutionary biology, and scalable machine learning architectures to produce robust, generalizable, and equitable typings. The discussion focuses on structural trade-offs in model design, data governance frameworks, computational infrastructure requirements, deployment strategies across diverse settings, sustainability of large-scale inference pipelines, fairness considerations in training and validation, and the policy implications for global genomic equity. We illustrate how explicit encoding of demographic histories and selective sweeps into model representations can reduce bias while improving predictive accuracy for underrepresented populations. The paper further examines the challenges of harmonizing heterogeneous long-read and short-read sequencing data across thousands of samples, the necessity of privacy-preserving architectures for sensitive genetic information, and the broader socio-technical infrastructure needed to support continuous learning from emerging population-level data. A case illustration based on the scalable framework for comprehensive typing from long-read data is used to contextualize these architectural decisions. The paper concludes by outlining a roadmap for future research that aligns technological development with ethical imperatives and global health priorities.