Dataset for structure-sensitive prediction of organic matter thermodynamics from quantum chemistry and machine learning

This dataset presents a complete and reproducible resource that links quantum chemical thermochemistry with machine learning to estimate the Gibbs free energy of oxidation per carbon for dissolved organic matter. It integrates three curated data tables, a companion archive of Gaussian log files, and three R scripts that regenerate all intermediate products and final predictions. The table scaled_energies_python_aq.csv contains thermochemistry information for molecular entries in water at 298 K using Gaussian software with B3LYP and the 6 311+G(2d,p) basis, the SMD solvent model, and a frequency scaling factor of 0.9614 (see methods section). The table balanced_data_aq_filtered.csv provides the refined 4030 molecules with reaction bookkeeping, normalised energy terms, and descriptors including elemental ratios, double bond equivalents, aromaticity index, and the nominal oxidation state of carbon. The file L.csv contains 1680 FTICR MS formula records with four sample intensity columns (SRFA, PLFA, ESFA, and SRNOM) and was generated in this study by processing raw MS data from machine L in (Hawkes et al., 2020) with the ICBM OCEAN (Merder et al., 2020) software to merge samples and assign molecular formulae to mass/intensity peaks. . Script one computes formation energetics from the log files and balances oxidation reactions. Script two trains a Random Forest model with cross validation to predict the normalised oxidation energy. Script three applies the model to the ICBM OCEAN (Merder et al., 2020) table and yields per formula predictions together with intensity weighted energy distributions aggregated by sample and compound class. All scripts can be run from the command line and save the R session details so that the analyses can be reproduced. Energies are reported in kJ molC-1 NB: This version is outdated (please use the newset version)