(169cq) Advancing Molecular Property Prediction and Chemical Reactivity Understanding through High-Throughput Quantum Chemistry and Graph Neural Networks

In this talk, we present our work in generating a one-of-a-kind foundational quantum mechanical (QM) dataset for thermochemistry, reaction kinetics, solvation free energies, and QM descriptors, generated using our open-source high-throughput computational chemistry workflow. This comprehensive dataset comprises over 345,000 open- and closed-shell reacting species, 200,000 transition states optimized at both the GFN2-XTB semi-empirical and ωB97X-D/def2-SVP density functional theory (DFT) levels of theory, DLPNO-CCSD(T)-F12d/def2-TZVP reaction barrier heights, and more than 100 million COSMO-RS solvation free energies in over 300 common solvents. The unparalleled scale and depth of this dataset can potentially lead to unprecedented insights into molecular properties and chemical reactivity.

By leveraging this extensive dataset and state-of-the-art Chemprop Directed Message Passing Neural Network (D-MPNN) models, we present promising preliminary results in accurately predicting molecular properties and chemical reactivity, such as the oxidative stability of diverse materials and solvent effects on reactivity. The generated dataset also paves the way for training foundational models, exploring active learning methods, and developing reactive neural network force fields. These advancements have far-reaching implications for further advancing autonomous chemical and material design, accelerating the discovery of novel materials with desired functional properties.

The combination of high-throughput quantum chemistry and advanced graph neural networks demonstrates the immense potential for accelerating materials discovery and optimization. By harnessing the power of this unique dataset and cutting-edge machine learning techniques, we pave the way for a deeper understanding of molecular properties and chemical reactivity, ultimately leading to more efficient and targeted design of functional materials for a wide range of applications.