WaSPred: A reliable AI-based water solubility predictor for small molecules

Abstract

A rapid and reliable evaluation of the aqueous solubility of small molecules is a hot topic for the scientific community and represents a field of particular interest in drug discovery. In fact, aqueous solubility significantly impacts various aspects that collectively influence a drug’s overall pharmacokinetics, including absorption, distribution and metabolism. For this reason, in silico approaches that provide fast and cost-effective solubility predictions, can serve as invaluable tools in the early stages of drug development. Although additional molecular features should be considered, accurate solubility predictions can help medicinal chemists rationally planning the synthesis of compounds more likely to exhibit desirable pharmacokinetic properties and in selecting the most promising candidates for further biological testing (e.g., cellular assays) from an initial pool of hit compounds with detected preliminary bioactivity. In this context, we herein report the development and evaluation of WaSPred, our AI-based water solubility predictor for small molecules. WaSPred not only showed high reliability in water solubility predictions performed on structurally heterogeneous compounds, belonging to multiple external datasets, but also demonstrated superior performance compared to a set of other commonly used water solubility predictors, thus confirming its state-of the-art robustness and its usefulness as an in silico approach for water solubility evaluations..