Project Grant R35GM166590

The National Institute of General Medical Sciences awarded the University of New Mexico Health Sciences Center $419,375 on August 25, 2026, under the Biomedical Research and Research Training program (CFDA 93.859) to develop two artificial intelligence models—Protein2Hypothesis and ProteinChem2Hypothesis—that translate protein and chemical sequences into human-readable functional descriptions and mechanistic explanations of protein function.

Project 1 will transform an existing Protein2Text model into Protein2Hypothesis, which predicts protein function, explains reasoning through mechanistic rationales, provides calibrated uncertainty estimates, and interprets the impact of coding variants including insertions, deletions, and frameshifts. This project will also produce DICE, a novel framework for instruction-guided protein design. Project 2 will develop ProteinChem2Hypothesis, a machine learning model that predicts interactions between protein and chemical representations. The team will leverage Azure cloud credits and compute allocations on the Frontier supercomputer at Oak Ridge National Laboratory to train large models and process massive datasets. The work addresses a gap created by the existence of 227 million known proteins and 111 million chemical compounds whose variants and interactions are too numerous to explore experimentally.

Performance occurs in Albuquerque, New Mexico. The period of performance runs through July 31, 2031.