Predicting Solubility of Chemical Compounds Computationally: An Ensemble Approach using QSAR Models

Abstract

Solubility is a property of a substance, a solute, to dissolve in a solvent. It is measured at equilibrium. It is defined as the maximum amount of solute that can be dissolved in a solvent at a given temperature. Solubility till now has always been physically computed. Various results are already available for the solubility of compounds. These results stand to vary as the results never show what experimental conditions were used. Value of solubility will vary even with the type of water(distilled/pH-buffered water) used in the experiment. Its area of application vary from drug designing, predicting the purification of salt, when will a solute form a precipitate, in qualitative analysis, in calculating remaining concentration after precipitation etc. This research deals with predicting the solubility of compounds using their physio chemical properties. SMILES and InChl structures have been chosen to train the data. Quantitative structure activity relationship models (QSAR models) have been used to study the relationship between the structure and the parameter of solubility. Various machine learning models were tuned and analysed. Top six models of machine learning were taken and ensembled. The best ensemble model was taken. The best outfit demonstrate was taken. K-fold cross-validation was utilized to check the power of the best model. The results show that, ensemble approaches can be successfully used for predicting the solubility. This research has the potential to reduce the burden on the pharmacologists by assisting them to calculate the solubility computationally.