Reaction mechanism prediction via automated reaction network analysis and chemical reactivity learning

Abstract

Despite remarkable advances in computational chemistry, prediction of reaction mechanism is still challenging, because investigating all possible reaction pathways is computationally prohibitive due to the high complexity of chemical space. In this thesis, we introduce the novel approach of automated chemical reaction prediction program named ACE-Reaction. ACE-Reaction constructs reaction network and extract the minimal network composed of only favorable reaction pathways from the complex chemical space. This method has been applied to various chemical reactions and verified its reliability and broad applicability. In addition, machine learning techniques and Monte Carlo tree search method are employed for chemical reactivity learning. It enables more efficient chemical reaction space exploration.