Discovering Low-Viscosity Molecules Using an Integrated Physics-Based Modeling, High-Throughput Screening and Active Learning Approach (1)— Screening from the GDB Database

The discovery of low-viscosity molecules is crucial for the development of next-generation batteries and capacitors. Large molecular libraries available in the literature provide a valuable resource for identifying promising candidates. In this study, we utilized the GDB database [1], one of the largest repositories of small molecules, to identify low-viscosity molecules. We employed and benchmarked molecular dynamics methods to accurately compute the dynamic properties without the need for synthesis or empirical testing, validating our calculations against experimental data. Physics-based simulations of viscosity included both Green-Kubo and Einstein-Helfand approaches allowing for robust calculation across the selected molecules. However, the number of molecules of interest from the GDB database is too large (several hundreds of thousands), making it impractical to identify promising candidates using purely physics-based models due to computational costs. Therefore, we implemented advanced machine learning (ML) techniques and smart selection approaches to dramatically reduce the number of physics-based calculations needed. By employing an active learning approach, we optimized the selection of molecules, enhancing the efficiency of the ML model while targeting low-viscosity candidates. Additionally, we computed the boiling points (BP) of the molecules using ML models trained on experimental BP data. As a result, we identified more than 100 molecules with viscosities less than 0.35 cP and BP above 80°C. We demonstrate that by integrating accurate physics-based models with advanced ML techniques, we can effectively identify top molecular candidates while significantly reducing computational costs.<br/><br/>[1] Enumeration of 166 billion organic small molecules in the chemical universe database GDB-17. Ruddigkeit Lars, van Deursen Ruud, Blum L. C.; Reymond J.-L. J. Chem. Inf. Model., 2012, 52, 2864-2875.