Data-Driven Mapping of Inorganic Chemical Space for the Design of Transition Metal Complexes and Metal-Organic Frameworks

Transition metal complexes have unique, tunable properties that make them attractive as catalysts or as molecular electronics but are challenging systems to model computationally owing to variable spin states, diverse metal-ligand bonding patterns and a lack of comparably large datasets to those that help train machine learning models on organic and bulk systems. This chapter reviews how these difficulties motivated the development of a series of unique representations, uncertainty-aware predictive models and an automated framework for machine-learning directed, first-principles powered materials design. In addition to choosing which complexes to simulate, these machine learning methods can integrate into, and accelerate, high-throughput first-principles screening through use of neural networks to efficiently initialize and control quantum chemical simulations, for example, detecting multireference character or avoiding calculations that cannot be converged. Finally, parallel efforts into understanding the diversity of metal-organic frameworks (MOFs) using these methods, supplemented by geometric information, are highlighted.