Decoding viral protein sequences by large language models.

Large language models (LLMs) for biological sequences are transforming computational biology, enabling a nuanced understanding of protein and nucleotide sequence data. Recent models, including ESM2, ESM3, AlphaGenome, Evo-1, and Evo-2, adapt natural language processing principles to the biological domain by learning high-dimensional hidden representations that capture evolutionary constraints, structural patterns, and functional motifs. This mini-review summarizes recent developments in devising and applying such models, emphasizing viral protein analysis. We highlight studies that have leveraged sequence-based LLMs in the protein domain (i.e. protein language models, or PLMs) for important application tasks such as viral protein annotation, variant effect prediction, and immune escape characterization. Additionally, we present a benchmark evaluation of these state-of-the-art protein language models to evaluate their core ability to capture evolutionary relationships between viral protein sequences. By discussing the opportunities and challenges of PLMs, the review outlines a road map for the potential application of LLMs in empowering virology research and pathogen surveillance.