Efficient learning of molecule properties with Graph Neural Networks and 3D molecule features

NLDL 2025 Conference Submission27 Authors

Keywords: Graph neural networks, molecules, graphs, deep learning

TL;DR: We present a new lightweight graph neural network for predicting molecule properties. Giving global 3D information about the molecule as input greatly helps the network. The model outperforms the state-of-the-art on 2 reference benchmark datasets.

Abstract:

Graph Neural Networks (GNNs) have emerged as a powerful tool in predicting molecular properties based on structural data. While GNNs excel in identifying local patterns within molecules, their ability to capture global properties remains limited due to inherent structural challenges such as oversmoothing.

We introduce an innovative GNN-based model that integrates global 3D molecular features with standard graph representations to enhance the prediction of molecular properties. The proposed model is evaluated using benchmark datasets ESOL and FreeSolv and it outperforms existing models. It demonstrates the crucial benefit of giving GNN models easy access to global information about the graph, in the context of applications to chemistry.

Additionally, the model's architecture allows for efficient training with relatively modest computational resources, making it practical for widespread application.

Submission Number: 27