Collaborative research: BIO-AI: modeling of large macromolecular systems at extra-long time scales using AI

The National Science Foundation Division of Molecular and Cellular Biosciences awarded $498,768 to the Stowers Institute For Medical Research on July 1, 2026, under the Biological Sciences program (CFDA 47.074) for a collaborative research project modeling large macromolecular systems using artificial intelligence.

The award funds development of an AI-driven approach to simulate protein assembly processes at physiologically relevant time scales in crowded cellular environments. The project combines structure-based modeling of molecular interactions, reinforcement learning theory, and experimental approaches to study protein aggregation. Research objectives include generating energy landscapes of macromolecular systems, developing and implementing an AI-based modeling protocol, and applying the approach to study emergent protein behavior, with specific focus on how crystallin proteins assemble to form a functioning lens rather than an amorphous aggregate. Graduate and undergraduate students will participate in projects applying AI to molecular biology problems. The work addresses a grand challenge in structural biology and carries implications for biotechnology methods used to exploit living systems for producing biological materials with industrial and medical applications.

Performance occurs at the Stowers Institute's location in Kansas City, Missouri, with a period of performance through June 30, 2029. The Stowers Institute is a tax-exempt nonprofit research organization conducting fundamental biomedical research, operating primarily as a prime recipient on federal project grants with federal funding concentrated within the National Institutes of Health.

Abstract

Knowledge of protein structures and their self-assembly in the crowded cell-like environment is essential for understanding how the molecular machines that enable biological processes are formed, and for our ability to modulate their formation when appropriate. This collaborative project combines highly complementary areas of expertise: structure-based modeling of molecular interactions, the theoretical foundation of reinforcement learning (a major direction in AI) and experimental approaches to study protein aggregation. Structural modeling of macromolecular assembly in the crowded environment of cell-size systems at physiologically relevant time scales is a grand challenge in structural biology. The project will develop the AI-driven approach needed to simulate physiologically relevant macromolecular processes and will study how crystallin proteins assemble to form a functioning lens rather than an amorphous aggregate. The implications of this work for advances in methods employed in the biotechnology sector are enormous?as this information will improve our ability to exploit living systems to produce biological materials with industrial/medical applications. Students at both the graduate and undergraduate level will participate in projects that apply AI to key problems in molecular biology. This important educational experience will significantly influence and expand their carrier options and choices. The objectives of the proposed research are: (1) generation of the energy landscape of the macromolecular system; (2) development and implementation of an AI-based modeling protocol; and (3) application to the study of the emergent protein behavior. The energy landscape of the macromolecular system is the foundation of the simulation approach, which samples the landscape in space and time. The accuracy of the landscape will be significantly improved by application of the AI-based predictions. The modeling protocol will be radically advanced to a new AI-driven paradigm. The emergent protein behavior will be investigated for the eye lens proteome. The lens crystallins likely operate through an entropic buffering mechanism. The aggregates promote their own growth by excluding volume that would otherwise be accessible to monomers, which creates an entropic drive for further aggregation. The hypothesis is that this mechanism is countered by the propensity of crystallins to form polydisperse native oligomers. The modeling approach, combined with wet-lab experiments, will provide a unique opportunity to test this hypothesis and to explore fundamental biology of protein homeostasis. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the foundation's intellectual merit and broader impacts review criteria.- subawards are not planned for this award.