Multi-Objective Design of Electrolyte Solvents Via Physics-Based Modeling and Reinforcement Learning

BP_Model_Training_Data.csv: This file contains the molecular structures used to train the boiling point (BP) prediction models. The dataset includes canonical SMILES strings for each molecule, along with the experimentally measured boiling point values. In addition, the file includes the full set of molecular descriptors computed and used in the development of both regression and classification models. These descriptors were generated using the DeepAutoQSAR (DAQ) platform and encompass RDKit-based descriptors. REINVENT_Generated_Molecules.zip files (eight files): This file includes the complete set of molecules generated across all eight REINVENT experiments: BC+VC, BC+VR, BR+VC, BR+VR, BC+VC+HR, BC+VR+HR, BR+VC+HR, and BR+VR+HR. Each entry includes the generated SMILES string and model-predicted values for boiling point, viscosity, and HOMO energy. This dataset provides a comprehensive view of the chemical space explored through reinforcement learning. Top_200_Validated_Candidates.csv: This file contains the 200 top-ranking molecules selected from the REINVENT-generated set, based on multi-objective criteria. For each molecule, the file provides the SMILES structure along with calculated properties from physics-based simulations. These include boiling point and viscosity values obtained from molecular dynamics (MD) simulations and HOMO energy values computed using density functional theory (DFT). These validated candidates represent the most promising molecules for further experimental consideration.