When novel or emerging pathogens (bacteria, viruses, parasites) are encountered, characterization of their interactions with human hosts currently requires weeks to months of experimental work, often yielding incomplete understanding. This capability gap limits rapid therapeutic response and countermeasure development. Recent advances in protein language models and large-scale protein-protein interaction (PPI) prediction make computational threat characterization feasible. This topic seeks to develop and validate an operationally deployable capability that can characterize any pathogen—naturally emerging, accidentally released, or engineered—from protein sequence data alone. The system must: (1) predict host-pathogen protein interactions with high accuracy across viral, bacterial, and parasitic pathogen classes; (2) demonstrate zero-shot prediction capability on previously unseen pathogens; (3) provide comprehensive functional annotation of both pathogen and host proteins; (4) generate ranked mechanistic hypotheses about infection pathways through automated analysis; and (5) complete core predictions within 15 minutes and full characterization reports within one hour on standard computing hardware. Proposers must demonstrate rigorous evaluation methods to ensure the system generalizes to unseen pathogens rather than memorizing training data. Performance must be benchmarked against established protein interaction databases and validated experimentally using standard binding assay techniques. The end-state capability enables rapid biological threat characterization to support medical countermeasure prioritization and force health protection.
No documents are available for this record yet.