Predicting Drug Solubility Using Different Machine Learning Methods -- Linear Regression Model with Extracted Chemical Features vs Graph Convolutional Neural Network

TL;DR

This study compares a linear regression model with RDKit-derived, hand-engineered features to a graph convolutional neural network (GCNN) for predicting molecular aqueous solubility across three public datasets. The GCNN generally achieves higher predictive accuracy, especially on larger and more diverse datasets, but at the cost of interpretability; the linear model provides transparent feature-level insights, such as the positive role of oxygen and the negative influence of most halogens. Including functional-group features improves the linear model's performance, underscoring the importance of chemical context beyond atom type alone. The authors discuss leveraging GCNNs alongside interpretable feature analysis and generative design to bridge prediction and drug design, enabling more efficient, informed high-throughput screening in pharmaceutical development.

Abstract

Predicting the solubility of given molecules remains crucial in the pharmaceutical industry. In this study, we revisited this extensively studied topic, leveraging the capabilities of contemporary computing resources. We employed two machine learning models: a linear regression model and a graph convolutional neural network (GCNN) model, using various experimental datasets. Both methods yielded reasonable predictions, with the GCNN model exhibiting the highest level of performance. However, the present GCNN model has limited interpretability while the linear regression model allows scientists for a greater in-depth analysis of the underlying factors through feature importance analysis, although more human inputs and evaluations on the overall dataset is required. From the perspective of chemistry, using the linear regression model, we elucidated the impact of individual atom species and functional groups on overall solubility, highlighting the significance of comprehending how chemical structure influences chemical properties in the drug development process. It is learned that introducing oxygen atoms can increase the solubility of organic molecules, while almost all other hetero atoms except oxygen and nitrogen tend to decrease solubility.

Figures (4)

• Figure 1: Configuration of the linear regression model (above) and GCNN model (below), using the tyrosine molecule as an example. The linear regression model relies on human-engineered features, including molecular weight (MW) and the count of functional groups, to predict experimental solubility (logS), whereas the GCNN utilizes features acquired via message across a graph.
• Figure 2: Parity plots for the Delaney, Huuskonen, and AqSolDB datasets using linear regression model (top) and GCNN model (bottom). Predictions are shown from the validation folds of 5-fold cross validation. Lines of best fit are shown in red.
• Figure 3: The linear regression weights of each type of atom feature for the Delaney dataset. Positive weights indicate features contributing to a relative increase in solubility, whereas negative weights indicate features which contribute to a relative decrease in solubility.
• Figure 4: The weights of each type of functional group feature in SMARTS notation. Positive weights indicate features contributing to a relative increase in solubility, whereas negative weights indicate features which contribute to a relative decrease in solubility.