Large language models improve annotation of viral proteins

Area of Science:

• Virology
• Bioinformatics
• Genomics

Background:

• Viral sequences in environmental samples are poorly annotated, hindering understanding of their role in microbial communities.
• Current sequence homology methods have limitations due to limited viral databases and sequence divergence.

Approach:

• Utilized protein language models (PLMs) to represent viral protein function.
• Applied PLMs to viral sequence annotation, focusing on protein family labeling and function identification.
• Expanded annotation of ocean virome viral protein sequences by 37%.

Key Points:

• PLM representations capture viral protein function beyond remote sequence homology.
• Identified a novel DNA editing protein family, indicating a new mobile element in marine picocyanobacteria.
• PLMs enhance detection of distant protein relationships and enable biological discovery.

Conclusions:

• Protein language models significantly improve viral protein annotation and functional characterization.
• This approach facilitates the discovery of novel viral elements and biological functions.
• Enhanced viral annotation is crucial for understanding viral impact on microbial ecosystems.