DMS/NIGMS 1: Mathematical and Computational Frameworks for Predicting Mutation-Induced Perturbations in Molecular Recognition

Protein mutations drive disease progression and therapeutic resistance by altering how proteins interact with other molecules in cells. A major scientific challenge is determining how small mutation-induced changes in protein structure and dynamics lead to large changes in biological function and treatment response. Current experimental approaches are often expensive, labor-intensive, and difficult to scale to the large number of possible mutations, while many computational methods lack the mathematical rigor and interpretability needed for reliable biological insight. This project will develop a new mathematical and artificial intelligence framework to predict and explain how mutations reshape biomolecular interactions associated with disease. Using estrogen receptor alpha (ERα), which plays a major role in endocrine therapy resistance in breast cancer, as an initial model system, the research will integrate topology, machine learning, and experimental biology to identify mutation-induced structural and functional changes linked to therapeutic response. The project will advance mathematical biology by creating transferable methods for analyzing complex biological systems and strengthening collaborations among mathematics, computation, and experimental science. The research will also support workforce development through interdisciplinary training of undergraduate and graduate students in applied mathematics, artificial intelligence, and molecular biology. Open-source software, curated datasets, and educational activities will expand access to advanced computational tools and help prepare a future workforce at the interface of mathematics and biology. This project will pursue three integrated objectives. First, it will develop a multiscale topological deep-learning (TDL) framework that incorporates physicochemical properties into topological descriptors to provide mathematically rigorous representations of mutation-induced perturbations in molecular recognition. The framework will capture both local residue-level perturbations and global conformational changes by identifying structural and biochemical signatures that accurately predict mutation effects. Although ERα mutations will serve as the primary model system, the framework will be designed for broad application to diverse protein–ligand and protein–protein interactions. Second, the project will generate benchmark datasets for model validation and generalization by integrating structural models, binding assays, and deep mutational scanning data into standardized and reproducible resources. The research will test whether harmonized multimodal datasets improve reproducibility and enable rigorous evaluation of predictive models across biological systems. Third, computational predictions will be experimentally validated using biochemical, biophysical, and cellular assays that measure ligand binding, coactivator and corepressor interactions, and functional responses associated with ERα mutations. Computational predictions and experimental results will be iteratively 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.