The First Frustrated Lewis Pairs Database: Machine Learning and Cheminformatics-Aided Prediction of Small Molecule Activation

Frustrated Lewis Pairs (FLPs) are a burgeoning field in chemistry, with novel FLPs and surprising new reactions continually being discovered. However, experimentally examining each FLP’s activity toward all possible small molecules (SMs) is impractical due to time constraints and potential dangers from toxic, flammable, or explosive SMs. Here, we developed the first open-access Frustrated Lewis pairs Database (FLPDB), with DFT-optimized atomic structures and computed binding free energies and electronic properties data for each FLP toward SMs, including H 2, CO 2, H 2 O, H 2 CO, HCOOH, and CH 3 OH, via high-throughput computational screening. Machine learning was employed to predict the small molecule binding free energies and identify the most important feature governing the binding free energies. Further integrating molecular fingerprints, the H 2 binding free energies can be predicted with a single feature─hydride affinity─for FLPs with molecular similarity scores ≥0.55. The FLPDB and the identified structure–activity relationships in this work will enable countless applications for novel FLP design, new reaction discovery, or toxic gas sensing and detection.