Efficiently navigating complex reaction networks with deep learning models – INVIGORATE

Complex reaction mechanisms are ubiquitous in chemistry. Many processes in transition-metal catalysis, polymer chemistry, enzyme catalysis and environmental chemistry involve a multitude of competing reaction pathways. Even in regular (organic) synthesis, the potential appearance of a plethora of (undesired) side reactions and products is of particular concern, since this limits the efficiency and applicability of many common – as well as prospective – transformations and catalytic systems. In order to analyze, control and/or steer the outcome of generic, complex reaction networks, all the relevant chemical compounds and associated elementary reactions need to be identified. Since the manual exploration of an entire mechanistic landscape quickly becomes intractable as the number of compounds present in a reaction medium increases, efficient automated exploration schemes are needed to achieve a detailed understanding of such networks at the atomistic scale. In the INVIGORATE project, we will use accurate (quantum chemistry augmented) machine learning models to efficiently navigate complex mechanistic landscapes, identifying potentially important pathways while discarding extremely unlikely branches of the network. By only explicitly computing pathways that are identified by these models as viable, one could make tremendous savings in terms of computational cost, facilitating more exhaustive explorations of more complex reaction mechanisms. Once a reaction network exploration platform has been developed, we intend to analyze various complex chemical systems with potential mechanistic ambiguity, mainly related to prebiotic chemistry.