Federal Grant Award Summary

Duke University's Office of Research Administration received a $375,996 Project Grant from the National Institute of General Medical Sciences (NIGMS) under the Biomedical Research and Research Training program (CFDA 93.859), effective July 1, 2026 through June 30, 2030. The award funds research to advance protein language models (PLMs)—computational tools that learn task-specific representations mapping protein sequences to functional outcomes. The research deliverables address three critical limitations in current PLMs: developing length-invariant generative representations that systematically explore insertions, deletions, and substitutions; introducing reverse distillation techniques to enable reliable scaling of large models through small-model guidance; and building modular aggregation layers that retain family-specific constraints while producing portable, task-aware sequence representations for classification and regression tasks.

The funded research will produce practical applications demonstrating these methodological advances across diverse protein engineering tasks, including optimizing mastoparan variants for antimicrobial potency with reduced cytotoxicity, improving fluorescent protein brightness through accurate mutational-effect modeling, and mapping epistatic rules governing viral immune escape in influenza. Upon completion, the project will deliver powerful, generalizable protein representations capable of handling complex mutations and producing well-calibrated predictions for protein design and prioritization across multiple biological domains.