Fig. 1
a Heat map of experimental partition coefficients as a function of carbon number. b Violin distribution plots of each elemental presence within the dataset. c Correlation matrix for feature-feature and feature-response correlations. The data set is highly populated from carbon numbers of 1–25 and \(LogP\) values of −2.5–5, this is region is expected to have the best performance. In addition to the quantity of data, compounds with chlorine, nitrogen and oxygen substitutions should lend to improved predictions as there are strong correlations to their partition coefficient