Predictive and therapeutic applications of protein language models

Area of Science:

• Artificial Intelligence
• Computational Biology
• Drug Discovery

Background:

• Protein language models (pLMs) are advanced AI tools trained on vast sequence data.
• These models offer powerful representational capabilities for complex biological problems.
• pLMs are transforming traditional approaches in therapeutic research.

Purpose of the Study:

• To provide a comprehensive review of pLMs and their implementations.
• To explore the utility of pLMs in drug discovery and therapeutic research.
• To discuss the current trends, applications, and future prospects of pLMs.

Main Methods:

• Systematic classification of pLMs based on architecture and data sources.
• Discussion of multimodal approaches integrating diverse biological information (co-evolutionary, structural, functional).
• Overview of domain-specific models for antibodies and T-cell receptors.

Main Results:

• pLMs demonstrate significant potential in various therapeutic applications.
• Key applications include mutation effect prediction, protein function prediction, and structure prediction.
• Multimodal and domain-specific models represent emerging trends.

Conclusions:

• pLMs are poised to significantly impact drug discovery and therapeutic development.
• Challenges remain in translating pLM capabilities into practical clinical applications.
• Future research should focus on overcoming these challenges for real-world disease contributions.