Establishing Protocols for Automated De Novo Design of Olefin Metathesis Catalysts

The increasing availability of high-performance computing resources and suitable cheminformatics methods is now bringing automated design of organometallic compounds within reach. As an example, we recently used the genetic algorithm implemented in the DENOPTIM (De Novo OPTimization of In/organic Molecules) package to search for highly active, ruthenium-based catalysts for olefin metathesis. The single design criterion (catalytic activity) was implemented in a correlation-based fitness score (figure of merit) relying on the known correlation between the stability of the metallacyclobutane intermediate and the barrier to productive metathesis. However, detailed follow-up studies revealed that the original fitness was inflexible and led to the prediction of sterically too encumbered catalysts. In the present work, we introduce a new and more flexible fitness that accounts for multiple properties: catalytic activity, resistance to decomposition via β-hydride elimination (β-HE), synthetic accessibility and thermodynamic stability, and on-the-fly detection of modeling artifacts. To ensure accurate predictions across a broad variety of ruthenium alkylidene complexes, the most critical components are computed from DFT-optimized transition-state models and do not rely on prediction models. The resulting protocol ranks the performance of well-known catalysts in agreement with experiment. Also in agreement with experiment is the fact that automated de novo design using this protocol identifies cyclic alkyl amino carbene (CAACs) as particularly promising ligands for productive catalysis. The present validated multi-component fitness protocol is currently being used in automated design of ruthenium olefin metathesis catalysts beyond the known NHC and CAAC paradigms.