Machine Learning-Powered Prediction of molecule Solubility: Paving the Way for environmental, and energy applications

¹ Mohammed V University in Rabat, ENSIAS, SSLAB, Rabat, Morocco
² Natural Resources and Sustainable Development laboratory, Faculty of Sciences, Ibn Tofail University, B.P 242, Kenitra, Morocco

^* Corresponding author: yassine.mouniane@uit.ac.ma

Abstract

Predicting aqueous solubility is pivotal for selecting materials in pharmaceuticals, environmental, and renewable energy fields. For instance, it plays a vital role in drug development and the design of chemical and synthetic routes. In the realm of Cheminformatics, the accurate prediction of molecule solubility is indispensable for drug discovery and development. Traditional methods often rely on labor-intensive experimental assays, presenting challenges in terms of time and cost. To address these limitations, this study leverages advanced machine learning techniques to predict molecule solubility with exceptional accuracy. Using the PyCaret library, a versatile low-code machine learning tool, we develop and evaluate a diverse set of linear regression models. Key performance metrics, including R², RMSLE, MAE, MSE, MAPE, and RMSE, are employed to assess model performance comprehensively. Through rigorous model comparison and evaluation, we identify the optimal model for predicting molecule solubility. Our findings not only demonstrate the efficacy of machine learning in Cheminformatics but also offer insights into the complex relationship between molecular features and solubility. This study contributes to the advancement of computational chemistry by bridging the gap between theory and practice. By elucidating the predictive capabilities of machine learning models, we pave the way for more efficient and cost-effective drug discovery processes.