Decoding viral protein sequences by large language models

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Large language models (LLMs) are increasingly adapted for biological sequence analysis, including protein and nucleotide data.
• Recent LLMs learn hidden representations capturing evolutionary, structural, and functional biological sequence information.
• These models, such as ESM2, ESM3, AlphaGenome, Evo-1, and Evo-2, are transforming computational biology.

Purpose of the Study:

• To review recent advancements in developing and applying LLMs for biological sequences, with a focus on viral protein analysis.
• To highlight the utility of protein language models (PLMs) in tasks like viral protein annotation, variant effect prediction, and immune escape characterization.
• To benchmark state-of-the-art PLMs for their ability to capture evolutionary relationships in viral protein sequences.

Main Methods:

• Literature review of recent developments in LLMs for biological sequences.
• Emphasis on sequence-based LLMs applied to viral protein analysis.
• Benchmark evaluation of selected state-of-the-art protein language models.

Main Results:

• LLMs, particularly PLMs, demonstrate significant potential in analyzing viral protein sequences.
• Applications include viral protein annotation, predicting variant effects, and characterizing immune escape.
• Benchmark results indicate the capability of PLMs in capturing evolutionary relationships.

Conclusions:

• LLMs and PLMs are powerful tools for understanding biological sequences, especially in virology.
• These models offer opportunities for enhanced viral protein analysis, pathogen surveillance, and outbreak response.
• Further research and development are needed to fully leverage the potential of PLMs in virology.