Pushing the limits of solubility prediction via quality-oriented data selection

Abstract

Accurate prediction of the solubility of chemical substances in solvents remains a challenge. The sparsity of high-quality solubility data is recognized as the biggest hurdle in the development of robust data-driven methods for practical use. Nonetheless, the effects of the quality and quantity of data on aqueous solubility predictions have not yet been scrutinized. In this study, the roles of the size and the quality of datasets on the performances of the solubility prediction models are unraveled, and the concepts of actual and observed performances are introduced. In an effort to curtail the gap between actual and observed performances, a quality-oriented data selection method, which evaluates the quality of data and extracts the most accurate part of it through statistical validation, is designed. Applying this method on the largest publicly available solubility database and using a consensus machine learning approach, a top-performing solubility prediction model is achieved.

RedDB, a Computational Database of Electroactive Molecules for Aqueous Redox Flow Batteries

“…The AqSolPred model, which was used for solubility predictions in the current work, had previously been validated on a benchmark solubility dataset 24 . The model has a Mean Absolute Error of 0.348 LogS, which is lower than the conventional cheminformatics and ML methods that are ordinarily used for the prediction of aqueous solubility of chemical species 13 .…”

“…It is an exemplary resource on quinone and aza-aromatic electroactive compounds as it contains several candidate molecules for batteries that are worthy of experimental investigation. The database contains comprehensive data that has been systematically collected by using the state-of-the-art computational procedures 1112 and data-driven methods 13 . Therefore, it's also useful for other applications beyond ARFBs for which the intriguing chemistry of these molecules matter.…”

“…The water solubility data of the compounds was built by using the Aqueous Solubility Prediction Model (AqSolPred v1.0) 13 . AqSolPred is a supervised and consensus ML model that was empowered by training on a large, curated, and reference aqueous solubility database, AqSolDB 22 .…”

“…All classical and quantum chemical calculations have been performed by using the SMSS 14 , which is a proprietary software package. The solubility predictions have been made by using the AqSolPred 13 , which is a freely accessible tool. In addition, the in-house developed Python scripts that have been used to parse the calculation outputs and to convert them into relational database formats, are openly accessible at https://github.com/ergroup/RedDB.…”