Data and trained models for predicting the stability and redox potentials of organic radicals

Part of an upcoming work on optimizing organic radical scaffolds for aqueous redox flow batteries.

• data/radical_db_redox_potentials.csv.gz contains the redox potentials (in V relative to SHE) calculated with Guassian 16c at the M06-2x/def2-tzvp level of theory in an implicit water solvent
• data/radical_db_spins_and_bv.csv.gz contains the parsed electron spins, normalized 'fractional spins' (ensuring all spin values are positive and sum to 1.0, assuming hydrogen spins are zero), and calculated buried volumes at a 3.5 Angstrom radius for all heavy atoms in the dataset. Atom index corresponds to the atom index assigned by RDKit after parsing the SMILES string.
• models/redox_model/ contains the trained weights for a GNN that predicts the oxidation and reduction potentials of a given radical input
• models/stability_model/ predicts the fractional spin and buried volumes on each atom in the given input radical, and can be used to derive the combined stability score.