Statistics and Bias-Free Sampling of Reaction Mechanisms from Reaction Network Models

Selection bias is inevitable in manually curated computational reaction databases but can have a significant impact on generalizability of quantum chemical methods and machine learning models derived from these data sets. Here, we propose quasireaction subgraphs as a discrete, graph-based representation of reaction mechanisms that has a well-defined associated probability space and admits a similarity function using graph kernels. Quasireaction subgraphs are thus well suited for constructing representative or diverse data sets of reactions. Quasireaction subgraphs are defined as subgraphs of a network of formal bond breaks and bond formations (transition network) composed of all shortest paths between reactant and product nodes. However, due to their purely geometric construction, they do not guarantee that the corresponding reaction mechanisms are thermodynamically and kinetically feasible. As a result, a binary classification of feasible (reaction subgraphs) and infeasible (non-reactive subgraphs) must be applied after sampling. In this paper, we describe the construction and properties of quasireaction subgraphs and characterize the statistics of quasireaction subgraphs from CHO transition networks with up to six nonhydrogen atoms. We explore their clustering using Weisfeiler--Lehman graph kernels.