(271a) Leveraging Experimental Data in Machine Learning Models to Accelerate the Discovery of New Materials and Catalysts

I will discuss our efforts to use machine learning (ML) to accelerate the computational tailoring and design of complex materials by leveraging experimental datasets. First, I will discuss metal-organic framework (MOF) materials and their application for catalysis as well as gas separations and storage. One limitation in a challenging materials space such as open shell transition metal chemistry present in the open metal sites of most catalytically active MOFs is that ML models and ML-accelerated high-throughput screening traditionally rely on density functional theory (DFT) for data generation, but DFT is both computationally demanding and prone to errors that limit its accuracy in predicting new MOFs. I will describe how we have curated a dataset of thousands of MOFs that have been experimentally synthesized and used this data to train ML models to predict experimentally reported measures of stability ranging from thermal stability to stability in water. I will describe how we have leveraged these models to then screen for mechanically stable materials as well as stable catalysts in the direct conversion of methane to methanol. I will also describe how we have used these models to accelerate the discovery of novel stable MOFs, creating a dataset of MOFs enriched with stability and diversity 1-2 orders of magnitude beyond what is typically included in most hypothetical MOF datasets. In the second half of my talk, I will discuss ways we have leveraged experimental data to build models on smaller data sets of molecular properties. Specifically, I will describe how we have aimed to develop a tool to design novel mechanophore constituents to mechanochemically reactive polymers. We have built an intuitive physical organic model that captures C–C bond reactivities under tensile force, by leveraging easy-to-compute molecular features in terms of force constants and reaction energies. I will describe how this model can accurately predict experimental transition forces. Finally, time-permitting, I will describe models and descriptors we have built for photochemical properties of transition metal complexes that are attractive design targets but challenging to compute with traditional modeling techniques.