Colloquium: Computer Science, Physics, Chemistry, and Biology: Predictive Modeling for Drug Discovery

Cray Distinguished Speaker Series
April 29, 2019 - 11:15am to 12:15pm
Ajay Jain
University of California at San Francisco
3-180 Keller Hall
Rui Kuang
Abstract: Approaches to computer-aided drug design span the spectrum from pure statistical regression approaches to pure physics-based simulation. The sins committed by researchers at each end of the spectrum are numerous but largely non-overlapping. Physicists often fall into a trap where they do not believe that their models have parameters, or that they are estimating parameters, and when they do, they may do so poorly. Computer scientists tend to estimate parameters well, especially in the context of machine learning approaches, but they often fall into a trap where they do not seek to understand and model molecular interactions in a physically realistic way. There is a middle path where careful parameter estimation based on data is done in the context of a functional model that embeds domain knowledge of protein/ligand interactions. This middle path will be explored in the context of predicting the bound configurations of drugs and drug-like molecules along with their binding affinities. The technical aspects of the work include object representation, non-linear function optimization, and multiple-instance machine learning. The presentation will be accessible to those with no prior knowledge of the wet side of drug discovery or of the relevant theoretical physics aspects.
Biography: Dr. Ajay N. Jain is appointed as Professor at the University of California, San Francisco, in the Department of Bioengineering and Therapeutic Sciences. Dr. Jain received his PhD in Computer Science in 1991 from Carnegie Mellon University working on computational machine learning methods. His undergraduate training was at the University of Minnesota, resulting in BS degrees in both Computer Science and in Biochemistry. He spent a number of years in the defense industry working on image processing and target recognition, concurrent with his academic training and for a brief period following his PhD. Beginning in 1992, Dr. Jain worked in a series of start-up biopharmaceutical companies on computational methods for structure-based drug design prior to joining UCSF in 1999. His research focus is in predictive computational modeling of biological systems, particularly as applied to drug discovery. His research encompasses state-of-the-art methods for computing small-molecule conformational ensembles, binding site modeling using machine learning, flexible molecular docking, ligand similarity, protein binding-site similarity, and predicting likely drug side-effects. With his long-term collaborator, Dr. Ann E. Cleves (Jain), direct application of the methods to drug discovery is carried out, both within the academic environment and with pharmaceutical industry partners.