Bourgeois Gadjagboui – Data Science
Hibbs Chemistry Conference room in the Science Building
Click here to join via Zoom
Title: Machine Learning Techniques for Molecular Properties Prediction
The solvation free energy measures the change in free energy that occurs when a solute molecule dissolves in a solvent. It is an important property in various biological, pharmacological, and chemical processes. Solvation free energy can be calculated using different theoretical and computational techniques such as molecular dynamics simulations and continuum solvent models. Prior to machine learning, solvation free energy prediction methods were classified into quantum mechanical, classical, and empirical categories. Quantum mechanical models treat molecular degrees of freedom at the quantum level, while classical models use molecular simulations. Empirical models include quantitative structure-property relations and linear solvation energy relationships. However, the use of implicit and explicit models has limitations that become a motivation for developing machine learning methods that can handle a larger number of solutes. In this study, we examine papers that have utilized machine learning methods to predict molecular properties. Our analysis reveals that molecular graph-based models are the most frequently utilized technique for estimating molecular properties. Additionally, we address the current limitations in the use of machine learning for predicting solvation free energy and pose some open questions.