GIPF electrostatic-surface descriptors of three experimental datasets from MD and conformer ensembles

Electrostatic-surface (GIPF) descriptors for three experimental datasets, each computed from three kinds of structure ensemble, as used in the paper Ensemble Molecular Surface Electrostatic Potential as Informative Features for Active Learning , available as a pre-print on ChemRxiv. For every molecule, the electrostatic potential of its partial charges is evaluated on the van der Waals surface and summarized by ten statistics (GIPF descriptors). These are averaged over: md: an explicit-solvent molecular dynamics trajectory (10 ns NVT production at 300 K; 80 frames from the last 8 ns); conf: an RDKit conformer ensemble, relaxed with MMFF94 in a dielectric equal to the solvent's (conformers within 10 kcal/mol of the lowest one; 8 to 300 per molecule); minconf: the lowest-energy conformer of that ensemble alone. All three use the same partial charges for a given molecule. Datasets: SurfPro: 1535 surfactants, GAFF2/AM1-BCC charges, water solvent (TIP3P; dielectric 78.4). Peptide Catalysis: 158 tripeptide catalysts of an annulation reaction (Wennemers & Denmark, ACS Cent. Sci. 2024), GAFF2/AM1-BCC charges, chloroform solvent (dielectric 4.81). Catalytic Peptide Fibers: 25 peptides that co-assemble into catalytic fibers, CGenFF charges, water (TIP3P; dielectric 78.4). Files: one CSV per dataset and ensemble type (9 in total), named <dataset>gipf<md|conf|minconf>.csv. Each row is one molecule. Each file has: the identifiers: a workdir key shared by the three files of a dataset, the SMILES as simulated, the formal charge, and the sequence for the peptide catalysts; the experimental targets as used in the paper, for SurfPro and Peptide Catalysis (the Fibers kinetics refer to peptide pairs and are not included); the number of structures averaged; the ensemble mean of each descriptor, and for md and conf its standard deviation. minconf has the 10 means only.