Viral G2P

Viral genomes contain extraordinary biological diversity, but most viral proteins still have no known function. Viral G2P develops language-model and probabilistic methods that connect viral genotype to phenotype at the scale of complete genomes and communities.

The research question

Rather than treating each protein in isolation, we model a virus as a structured collection of proteins. This makes it possible to ask which proteins consistently occur together, which environmental contexts shape those relationships, and how protein language models can recover functional similarity even when conventional annotations fail.

Current directions

• Fine-tuning protein language models for underrepresented viral proteomes
• Extending protein representations to genomic scale with biologically induced sparse attention
• Inferring viral gene function through interpretable soft alignments
• Connecting protein composition to viral phenotype and ecology

This work is developed with collaborators in viral ecology, microbiology, and computational biology.