CAREER: Physics-Constrained Modeling of Molecular Texts, Graphs, and Images for Deciphering Protein-Protein Interactions

Texas A&M Engineering Experiment Station, College Station TX

Investigators

Abstract

Proteins are essential parts of biological systems that often function through interactions. Toward understanding and engineering biological systems, data are rapidly accumulating on what proteins and what protein-protein interactions (PPIs) are present in such systems, but a major barrier remains as knowledge is limited on how proteins interact in 3-dimensional (3D) space. This project is designed to help fill the knowledge gap by developing computational methods that predict mechanism-revealing 3D structures formed by PPIs. While developing such methods, a data-focused yet physics-rationalized approach will be pursued, which is expected to advance the state of the knowledge across natural science and artificial intelligence. The outcome of the project will facilitate deciphering and engineering genome-wide PPIs for wide applications such as novel therapeutics, clean energy, and smart materials. The project is also designed with educational activities to promote the awareness, participation, training, and communication of data-driven science discovery for students, educators, domain scientists, and general public. The highly interdisciplinary research and education activities will be integrated to foster a diverse globally-competitive workforce, including historically underrepresented groups, to be ready for the era of big data. The research goal of this project is to advance the state of the art for structural PPI prediction and re-think and tackle the problem as explaining how pairs of proteins, represented in various data forms such as texts, graphs, or images, interact under governing physics. In pursuit of the goal, the research objectives of the project involve three levels of PPI structural prediction of increasing resolutions and challenges: residue-level contact maps, residue-level distance distributions, and atom-level 3D structures. Initiated by these objectives, novel machine learning algorithms will be developed and contribute to foundational algorithm research, including the effective integration and learning from heterogeneous data as well as the flexible representation and incorporation of domain knowledge. Such advance in foundational algorithm research will expand the applicability of PPI structural prediction to genome-scale and learn physical principles underlying diverse PPIs rather than “memorizing” patterns in similar PPIs. Moreover, such methodological advance is expected to impact broad application fields beyond PPI structural prediction. The proposed research is integrated with an educational plan by feeding research results and trained personnel to multi-scale education and outreach activities, involving educated students in research, and engaging general public in citizen science. New curricular and co-curricular activities will be developed to enhance the accessibility to interdisciplinary data-science training for a diverse student body and domain scientists. Also, multi-level outreach activities in collaboration with existing programs will be used to foster the awareness of and interest in interdisciplinary data science among diverse middle- and high-school students as well as the general public. 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.