Explainable Solvation Free Energy Prediction Combining Graph Neural Networks with Chemical Intuition

Abstract

The prediction of a molecule’s solvation Gibbs free (ΔG solv) energy in a given solvent is an important task which has traditionally been carried out via quantum chemical continuum methods or force field-based molecular simulations. Machine learning (ML) and graph neural networks in particular have emerged as powerful techniques for elucidating structure–property relationships. This work presents a graph neural network (GNN) for the prediction of ΔG solv which, in addition to encoding typical atom and bond-level features, incorporates chemically intuitive, solvation-relevant parameters into the featurization process: semiempirical partial atomic charges and solvent dielectric constant. Solute–solvent interactions are included via an interaction map layer which can be visualized to examine solubility-enhancing or -decreasing interactions learnt by the model. On a test set of small organic molecules, our GNN predicts ΔG solv in water and cyclohexane with an accuracy comparable to polarizable and ab initio generated force field methods [mean absolute error (MAE) = 0.4 and 0.2 kcal mol–1, respectively], without the need for any molecular simulation. For the FreeSolv data set of hydration free energies, the test MAE is 0.7 kcal mol–1. Interpretability and applicability of the model is highlighted through several examples including rationalizing the increased solubility of modified diaminoanthraquinones in organic solvents. The clear explanations afforded by our GNN allow for easy understanding of the model’s predictions, giving the experimental chemist confidence in employing ML models toward more optimized synthetic routes.

Machine Learning Methods for Small Data Challenges in Molecular Science

“…The success of GNN in many domains such as recommender systems, , computer vision, and NLP , is attributed in part to its effectiveness in extracting latent representations from Euclidean data. However, as data are increasingly represented in the form of graphs, including non-Euclidean domains such as e-commerce, , chemistry, , and citation networks, , there is a growing need for GNNs. Additionally, molecular property prediction is a popular application of GNNs, as molecules can be represented as topological graphs, with atoms as nodes and bonds as edges.…”

Comment on ‘Physics-based representations for machine learning properties of chemical reactions’

“…If working with solvated systems, some papers claim that solvent-based splits-in which the model sees molecules or reactions in some solvents during training and then performance is tested with the same molecules or reactions in new solvents-can measure extrapolation [119]. However, given that ∆G solvation values often have relatively small magnitudes [120][121][122], solvent-based splits are likely not as challenging as they would appear, and it is likely that a trivial baseline model would perform only slightly worse; in contrast, solute-based splits represent a harder task [123][124][125].…”

Ground truth explanation dataset for chemical property prediction on molecular graphs

“…A wide range of machine learning (ML) approaches allows for explaining the chemistry of molecules, attributing which parts of the molecules are responsible for the chemical property of interest [1][2][3][4][5][6][7][8][9][10][11][12][13][14][15][16][17][18][19] , and lessening the black box challenge of machine learning 20,21 . Typical explainable ML approaches that provide atomwise attribution include dummy atoms 22 , classification of atoms by chemical intuition 23 , regression models 24 , graph neural network (GNN) attributions [25][26][27][28] with gradients 29 , perturbations 30 , decompositions 31 , and surrogates 32 .…”