GENERATIVE AI FOR PROTEIN MODELING: SAMPLING PROTEIN CONFORMATION SPACE

Protein modeling has benefitted greatly from machine learning methods over the years,particularly in well-defined prediction tasks attempting to connect the chemical characteristics and makeup of protein molecules with their biological activities and function in the cell. Many decades’ worth of experimental, computational, and theoretical studies, however, have informed us both on the rich set of activities a protein molecule can carry out in the cell, as well as on the intrinsic structural plasticity that allows a protein molecule to interact with diverse molecular partners. Accounting for such plasticity by sampling, for instance, the potentially rich space of conformations of a protein molecule has motivated much computational research over the years and remains a challenging problem. In this dissertation, we build over the generative AI platform, harnessing the growing sophistication of generative deep learning to address protein structure modeling at increasing complexity. Specifically, this dissertation proposes, implements, and rigorously evaluates deep generative models of various architectures for sampling protein conformation space. The work presented in this dissertation work advances protein structure modeling and, more broadly, generative AI for science.

Author

• RAHMAN, TASEEF

Identifier

hdl:1920/13683

OAI identifier oai:identifier

oai:MARS:1920/13683