We combine modern big quantum chemistry data sets with machine learning techniques to tackle an outstanding challenge in experimental chemistry: The development and application of QML models capable of providing accurate predictions of chemical reactions in real-time. Accomplishing this goal would represent a major advancement for chemistry, equipping bench chemists all over the world with a tool to interactively and reliably plan their reactions prior to experimentation. The atomistic simulation community will also profit from these developments as substantially larger sets of reaction profiles will become accessible for subsequent analysis. On a conceptual level, reaching real-time speed and chemical accuracy for chemical reactions implies a profound deepening of our understanding of chemistry, and may even lead to the discovery of new chemical reactions or catalysts. It is well possible that hitherto unknown rules, akin to Hammett-relationships, will be discovered as a result of the proposed work. If successful our work will be of relevance wherever chemical reactions occur, i.e. also in the biomolecular or materials sciences. Work performed from the beginning of the project to the end of the period covered by this report includes the development of quantum machine learning models for Hammett-relationships, van der Waals corrections, reaction barriers, free energies of solvation, as well as quantum chemistry data-sets covering (a) substitution and elimination reactions, and (b) reactive open-shell species (carbenes). A novel symmetry relationship in chemical compound space, dubbed ‘alchemical chirality’, has also been discovered, and leveraged to derive novel baselines for Delta Machine Learning models of quantum properties applicable throughout chemical space. Overall progress has pushed the frontier of science and includes quantum machine learning models with atomistic details in reactive processes, enabling unprecedented predictive power when making chemical reactivity estimates in real-time.
