Modeling and Predicting Catalytic Organic Reactions

Bio: Peng Liu is the Richard King Mellon Chair of Chemistry at the University of Pittsburgh. He earned his B.S. from Peking University in 2003 and his M.Sc. from the University of Guelph in 2006. Following the completion of his Ph.D. and postdoctoral research at UCLA, he joined the University of Pittsburgh faculty in 2014. He was promoted to Associate Professor in 2019, Full Professor in 2022, and the Richard King Mellon Chair in 2025. Professor Liu’s research accomplishments have been recognized with numerous honors, including the NSF CAREER Award (2017), the Journal of Physical Organic Chemistry Award for Early Excellence (2018), the NIH Maximizing Investigators' Research Award (2018), the Innovation in Education Award (2020), and the Chancellor’s Distinguished Research Award (2020).

Abstract: Rational catalyst design in transition metal catalysis and biocatalysis remains largely empirical due to the mechanistic complexity of multistep catalytic cycles and intricate catalyst–substrate interactions. This talk describes our group’s computational workflow for modeling, analyzing, and predicting catalytic organic reactions, organized around two themes: complexity and generality.

The first part addresses complexity through computational approaches that go beyond conventional DFT. Energy decomposition analysis (EDA) quantitatively dissects the non-covalent interactions governing catalyst reactivity and selectivity. We further demonstrate through ab initio molecular dynamics (AIMD) simulations with explicit solvent that dynamic flexibility effects fundamentally alter operative reaction pathways and enantioinduction mechanisms. These insights extend to enantioselectivity prediction in dynamic environments, including dirhodium-catalyzed C–H functionalization and enantioselective enzyme catalysis.

The second part addresses generality—translating mechanistic insights into broadly applicable predictive tools. Mechanistically guided regression models use physically meaningful descriptors from transition state analysis to predict reactivity across diverse catalyst and substrate space, exemplified by Cu-catalyzed atom-transfer radical polymerization (Cu-ATRP). We also present two large-scale descriptor databases—HArD (31,500+ heteroaryl groups) and RAPID (1M+ radicals)—augmented by graph neural network models for rapid descriptor prediction, with applications to reactivity predictions for covalent inhibitors, C–H functionalization, and polymer upcycling.

The talk concludes with an outlook on merging multiscale simulations with machine learning to predict reactivity and selectivity in complex catalytic systems, including de novo enzyme design, representing a shift toward physics-informed, data-driven catalyst discovery.