Evolution-based understanding and design of proteins

The National Institute of General Medical Sciences awarded The University of Chicago $487,525 on March 1, 2026, under the Biomedical Research and Research Training program (CFDA 93.859) to develop evolution-based understanding and design of proteins through generative AI models and experimental validation.

The work centers on deducing design principles by which high-performance, robustness, and adaptability are realized in protein evolution. The recipient will develop a generative model enabling text-prompted protein engineering of functional artificial proteins, conduct experiments in protein dynamics and evolution to conceptually understand information encoded by generative models, and deliver experimentally validated models for protein engineering alongside a technology platform for model-driven approaches to biological systems. The research unifies modern AI approaches with experimental methods and will serve as a training environment for quantitative scientists in biology.

Performance occurs in Chicago, Illinois through December 31, 2030. This is a Project Grant assistance award.

Abstract

Evolution builds biological systems that can self-assemble and display complex information and material processing capabilities that rival or exceed the performance of man-made systems. For example, protein molecules fold spontaneously into ordered structures and exhibit the ability for specific binding, catalysis, signal transmission, and allosteric regulation. They can do all these things while remaining robust to random variation and adaptive to changing conditions of selection in the environment. The central goal of our work is to deduce the design principles by which high-performance, robustness, and adaptability are realized in the process of evolution. Past work using sequence-based statistical models give US clear hypotheses to test, and new technologies for deep generative models, protein dynamics, and forward evolution experiments open up for the first time a definitive opportunity to test the hypotheses. In this proposal, we describe a new generative model that enables text-prompted protein engineering of functional artificial proteins with great diversity. We also describe new experiments for dynamics and evolution that allow US to conceptually understand the information encoded by generative models. Since this work represents a unification of modern AI approaches with cutting- edge experimental methods, the proposed studies promise to provide an excellent training environment for young quantitative scientists working in biology. Overall, the outcomes of this research program will be experimentally validated models for protein engineering, new concepts of how proteins work and evolve, and a technology platform that can empower the scientific community to advance model-driven approaches for biological systems.