Hybrid, Interpretable Machine Learning for Thermodynamic Property Estimation using Grammar2vec for Molecular Representation

Property prediction models have been developed for several decades with varying degrees of performance and complexity, from the group contribution-based methods to molecular simulations-based methods. An interesting issue in this area is finding an appropriate representation molecules inherently suited property modeling problem. Here, we propose Grammar2vec, a SMILES grammar-based framework generating dense, numeric representations. Grammar2vec embeds structural information contained grammar rules underlying string representations molecules. We use build machine learning-based estimating normal boiling point (Tb) critical temperature (Tc) benchmark their against popularly used contribution (GC)-based To ensure interpretability ML model, perform Shapley values-based analysis estimate feature importance simplify (or prune) trained model.