ESOL:  Estimating Aqueous Solubility Directly from Molecular Structure

Key Points

• To develop a simple, structure-based computational model for estimating aqueous solubility without requiring experimental physical property inputs.
• Trained a linear regression model using measured aqueous solubility data from 2,874 diverse chemical compounds against nine molecular parameters.
• Assessed predictive accuracy across three independent validation sets of medicinal and agrochemical-sized compounds against the General Solubility Equation.
• Identified calculated logP, molecular weight, proportion of aromatic heavy atoms, and count of rotatable bonds as the primary structural predictors of solubility.
• Predicted aqueous solubility within a factor of 5 to 8 of experimental values across three validation sets, matching the performance of standard empirical equations.