Studying the Deep Evolution of Viruses in the Era of Artificial Intelligence Structure Prediction

High mutation rates erode viral sequence similarity, obscuring deep evolutionary history. While protein structure is far more conserved than sequence, its use in evolutionary studies has historically been bottlenecked by experimental determination. The recent revolution in artificial intelligence (AI) structure prediction has fundamentally changed this, enabling the rapid generation of millions of viral protein structures. This review examines the effect of AI-based protein structure prediction methods on our understanding of deep viral evolution. We describe the strengths and limitations of protein structure prediction and consider the questions it can be used to address: illuminating viral dark matter in metagenomic datasets, resolving high-level taxonomy for orphan lineages, and inferring function for divergent proteins. Furthermore, we assess the emerging field of structural phylogenetics, exploring the theoretical and practical challenges of integrating structure and sequence to reconstruct ancient evolutionary events. We conclude that despite remaining challenges, systematic structure prediction will extend our exploration of deep evolution across the virosphere.