Contrastive Dual-Interaction Graph Neural Network for Molecular Property Prediction

TL;DR

DIG-Mol is introduced, a novel self-supervised graph neural network framework for molecular property prediction that leverages the power of contrast learning with dual interaction mechanisms and unique molecular graph enhancement strategies to efficiently improve molecular characterization.

Abstract

Molecular property prediction is a key component of AI-driven drug discovery and molecular characterization learning. Despite recent advances, existing methods still face challenges such as limited ability to generalize, and inadequate representation of learning from unlabeled data, especially for tasks specific to molecular structures. To address these limitations, we introduce DIG-Mol, a novel self-supervised graph neural network framework for molecular property prediction. This architecture leverages the power of contrast learning with dual interaction mechanisms and unique molecular graph enhancement strategies. DIG-Mol integrates a momentum distillation network with two interconnected networks to efficiently improve molecular characterization. The framework's ability to extract key information about molecular structure and higher-order semantics is supported by minimizing loss of contrast. We have established DIG-Mol's state-of-the-art performance through extensive experimental evaluation in a variety of molecular property prediction tasks. In addition to demonstrating superior transferability in a small number of learning scenarios, our visualizations highlight DIG-Mol's enhanced interpretability and representation capabilities. These findings confirm the effectiveness of our approach in overcoming challenges faced by traditional methods and mark a significant advance in molecular property prediction.

Figures (9)

• Figure 1: The overarching architecture of DIG-Mol, a two-stage contrastive learning framework. Figure 1. depicts the primary components of the framework, encompassing five key submodules: A. Overview of DIG-Mol pre-training process. B. Overview of DIG-Mol finetuning stage. C. Potential applications of molecular property prediction. D. Computational logic of various contrastive losses in DIG-Mol. E. Legends of data streams and other contents in the DIG-Mol architecture.
• Figure 2: Graph augmentation operations with '1-methoxy-2-methylbenzene ' as the target. Node masking (Fig 2A), set the features of corresponding nodes(atoms) to zero in the feature matrix. Unidirectional edge deletion (Fig 2B), delete the one-way message passing(edges) in the adjacent matrix.
• Figure 3: The ROC curves for different classification datasets. The different colored curves represent different subtasks. Please note that the ROC-AUCs plot may not exactly correspond to the best results in the figure legends.
• Figure 4: Visualization of spatial localization of molecular representations in the BBBP dataset. The left column displays the initial representations, while the right column showcases pretrained representations
• Figure 5: DIG-Mol Visualizations for Exploring Chemical Interpretation using the FreeSolv Dataset. The 2-D visualization depicts molecular representations generated by pre-trained DIG-Mol. The 3-D visualization illustrates molecular representations generated by DIG-Mol during the fine-tuning phase, with the height axis corresponding to the negative values of molecular hydration free energy. Circumambient molecules represent key molecules within hydrophilic and hydrophobic clusters, delineated by dashed circular outlines in the visualization space.