Protein Language Models in Virology: A Review of Advances and Applications

Area of Science:

• Virology
• Bioinformatics
• Artificial Intelligence

Background:

• Traditional viral sequence analysis relies on homology, limiting the study of divergent sequences.
• Protein Language Models (PLMs) offer a novel approach for functional analysis independent of sequence similarity.

Purpose of the Study:

• To review the diverse architectures of PLMs, including sequence encoders, structure-aware models, and generative models.
• To assess the current applications of PLMs in virology, such as protein structure resolution, functional classification, host factor identification, and therapeutic antibody optimization.
• To discuss the limitations of existing PLMs and explore future directions for AI in precision virology.

Main Methods:

• Review of existing literature on Protein Language Models (PLMs).
• Categorization of PLM architectures (sequence encoders, structure-aware, generative).
• Assessment of PLM applications across various virological research areas.

Main Results:

• PLMs can analyze divergent viral sequences without homology alignment, enabling functional insights.
• Applications span orphan protein structure resolution, virosphere-wide functional classification, host factor identification, and therapeutic antibody optimization.
• Current limitations include challenges in interpretability and data representation.

Conclusions:

• PLMs are powerful tools for advancing precision virology by enabling functional analysis of diverse viral sequences.
• Future trends involve multimodal integration and the "dry-wet" experimental loop to accelerate AI adoption.
• Addressing limitations in interpretability and data representation is crucial for future PLM development.