QSPR models for estimating retention in HPLC with the p solute polarity parameter based on the Monte Carlo method

Quantitative structure–property relationship (QSPR) models are built for the set of 233 very different organic chemical compounds for the estimation of the solute polarity parameter p. QSPR models for the solute polarity parameter p are calculated with optimal descriptors based on a SMILES notation, and all models were developed using the Monte Carlo method where the end point is threaded as a random event. Models were prepared in accordance with the OECD principles and recommendations. Three random splits into the training, test, and validation sets were examined. The statistical quality of all build models was very good. The best calculated model had the following statistical parameters: for the training set r 2 = 0.9493, q 2 = 0.9479, s = 0.285, F = 2564; r 2 = 0.9608, q 2 = 0.9561, s = 0.249, F = 1102 for the test set; and r 2 = 0.9418 and q 2 = 0.9349 for the validation set. Structural indicators (alerts) defined as molecular fragments for the increase/decrease in the solute polarity parameter p were defined.