Viro3D: a comprehensive database of virus protein structure predictions

Area of Science:

• Molecular Biology
• Virology
• Structural Biology
• Bioinformatics

Background:

• Viral genetic sequences diverge rapidly due to mutations and host adaptations, complicating evolutionary and functional studies.
• Protein structure is more conserved than sequence, offering a more stable basis for phylogenetic analysis and functional annotation.
• Existing protein structure databases significantly underrepresent viral proteomes.

Purpose of the Study:

• To leverage machine learning for large-scale prediction of viral protein structures.
• To create a comprehensive database of viral protein structures for the research community.
• To analyze viral evolution and function using predicted structural data.

Main Methods:

• Utilized AlphaFold2-ColabFold and ESMFold for structure prediction.
• Generated structural models for over 85,000 proteins from more than 4400 viruses.
• Expanded viral protein structure coverage by 30-fold compared to experimentally determined structures.

Main Results:

• Created the Viro3D database, significantly enhancing viral protein structure representation.
• Mapped protein form and function across human and animal viruses.
• Investigated the evolutionary history of viral class-I fusion glycoproteins, including potential origins of coronavirus spike proteins.

Conclusions:

• Machine learning-driven structure prediction revolutionizes viral proteome analysis.
• The Viro3D database provides a valuable resource for molecular virology research.
• Structure-informed insights facilitate the design of novel antiviral therapies and vaccines.