Efficient epistasis inference reveals constraints on HIV-1 evolution

Epistasis, the non-additive effects of mutations that alter phenotypic traits, is common in rapidly evolving populations like microbes and viruses. Detecting and inferring these epistatic interactions is essential for understanding how epistasis arises through evolution and influences evolutionary trajectories. The increasing availability of high-throughput genetic sequencing offers a unique opportunity to study epistasis in rapidly evolving pathogens. However, inferring epistasis from temporal genetic data is challenging due to genetic linkage, where mutations appear correlated not because the concurrence alters phenotypes but due to shared evolutionary history. Disentangling genetic linkage from underlying epistasis is crucial for accurate inference. While recent statistical models have addressed genetic linkage, they are limited to short sequences and are impractical for real-world genetic sequences.

Here, we propose an efficient framework for epistasis inference that accounts for genetic linkage and is scalable to large genetic datasets. Using this method, we infer epistasis from intrahost HIV-1 evolution, stereotypical complex evolutionary processes. We will describe further the epistatic patterns identified and discuss their biological significance.