Las Vegas algorithm in the prediction of intrinsic solubility of drug-like compounds

Abstract

A randomized algorithm that always succeeds in producing a correct output, but whose running time depends on random events is known as a Las Vegas algorithm. In this study, the Las Vegas algorithm aimed to improve QSPR models of intrinsic solubility of drug-like compounds obtained by the Monte Carlo method. Corresponding computational experiments were carried out with the CORAL software. The developed QSPR models were rigorously validated using a battery of statistical parameters, demonstrating excellent predictive ability and robustness. It has been shown, that the Las Vegas algorithm is a suitable way to improve the predictive potential of models obtained with the Monte Carlo technique. Additionally, the study identified key molecular fragments derived from the SMILES notation descriptors that influence the intrinsic solubility (increase or decrease). Overall, this work underscores the efficacy of the Monte Carlo method optimization with applied Las Vegas algorithm in constructing conformation-independent QSPR models with strong predictive power for prediction of intrinsic solubility of drug-like compounds.