Could graph neural networks learn better molecular representation for drug discovery? A comparison study of descriptor-based and graph-based models

TLDR

Graph neural networks have emerged as a promising approach for molecular property prediction, often outperforming traditional descriptor-based methods. This study evaluates and compares the predictive performance and computational efficiency of eight machine learning algorithms—four descriptor-based (SVM, XGBoost, RF, DNN) and four graph-based (GCN, GAT, MPNN, Attentive FP)—across 11 public datasets covering diverse property endpoints. The authors benchmarked these algorithms on the datasets, assessed accuracy and training time, applied SHAP for model interpretation, and tested the models in virtual screening against HIV targets. Descriptor-based models, particularly SVM for regression and RF/XGBoost for classification, generally achieved higher accuracy and faster training than graph-based models, with some graph methods excelling on larger or multi-task datasets, and overall the results support the continued use of off-the-shelf descriptor-based models for accurate, efficient, and interpretable predictions.

Abstract

Abstract Graph neural networks (GNN) has been considered as an attractive modelling method for molecular property prediction, and numerous studies have shown that GNN could yield more promising results than traditional descriptor-based methods. In this study, based on 11 public datasets covering various property endpoints, the predictive capacity and computational efficiency of the prediction models developed by eight machine learning (ML) algorithms, including four descriptor-based models (SVM, XGBoost, RF and DNN) and four graph-based models (GCN, GAT, MPNN and Attentive FP), were extensively tested and compared. The results demonstrate that on average the descriptor-based models outperform the graph-based models in terms of prediction accuracy and computational efficiency. SVM generally achieves the best predictions for the regression tasks. Both RF and XGBoost can achieve reliable predictions for the classification tasks, and some of the graph-based models, such as Attentive FP and GCN, can yield outstanding performance for a fraction of larger or multi-task datasets. In terms of computational cost, XGBoost and RF are the two most efficient algorithms and only need a few seconds to train a model even for a large dataset. The model interpretations by the SHAP method can effectively explore the established domain knowledge for the descriptor-based models. Finally, we explored use of these models for virtual screening (VS) towards HIV and demonstrated that different ML algorithms offer diverse VS profiles. All in all, we believe that the off-the-shelf descriptor-based models still can be directly employed to accurately predict various chemical endpoints with excellent computability and interpretability.