Machine Learning Seminar by Dr. Xia Ning – Deep Generative Models for Molecule Optimization

Dr. Xia Ning is an Associate Professor in the Biomedical Informatics Department, and the Computer Science and Engineering Department, The Ohio State University. She received her Ph.D. in Computer Science and Engineering from the University of Minnesota, Twin Cities, in 2012. Ning’s research is on Artificial Intelligence (AI) and Machine Learning (ML) with applications in drug discovery, health care, and e-commerce. Her research has been highly interdisciplinary. Her lab develops innovative and effective AI models and computational methods to derive knowledge from heterogeneous big data, conduct modeling, ranking, classification, prediction, and generation, etc., and ultimately solve critical and real high-impact problems. Specific applications include new molecule generation and drug candidate prioritization for drug discovery, cancer drug selection for precision medicine, drug repurposing for Alzheimer’s Disease, and information retrieval from electronic medical records. Ning won the 10-Years-Highest-Impact-Paper Award, IEEE International Conference on Data Mining (ICDM) in 2020.

Molecule optimization is a critical step in drug development to improve desired properties of drug candidates through chemical modification. In this talk, I will present a novel deep generative model Modof over molecular graphs for molecule optimization. We developed Modof leveraging the most advanced deep learning approaches that enable profound molecule structure representation learning and new molecule generation through sampling from molecule representations and encoding.  Following the rationale of fragment-based drug design, Modof modifies a given molecule by predicting a single site of disconnection at the molecule and the removal and/or addition of fragments at that site. A pipeline of multiple, identical Modof models is implemented into Modof-pipe to optimize molecules at multiple disconnection sites. Here we show that Modof-pipe can retain major molecular scaffolds, allow controls over intermediate optimization steps, and better constrain molecule similarities. Modof-pipe outperforms the state-of-the-art methods on benchmark datasets, with a 121.0% property improvement without molecular similarity constraints, and 82.0% and 10.6% improvement if the optimized molecules are at least 0.2 and 0.4 similar to those before optimization, respectively.