[Paper Review] A Hypergraph Convolutional Neural Network for Molecular Properties Prediction using Functional Group.

Fangying Chen, Junyoung Park+1 more|arXiv (Cornell University)|Jun 2, 2021

Computational Drug Discovery Methods10 references4 citations

TL;DR

This paper proposes MolHGCN, a hypergraph convolutional network that integrates atom and functional group information to improve molecular property prediction. By modeling multi-node interactions via hypergraphs and using a two-stage message passing mechanism, MolHGCN achieves superior performance over baseline methods on multiple benchmark datasets, particularly due to enhanced latent space separability from functional group integration.

ABSTRACT

We propose a Molecular Hypergraph Convolutional Network (MolHGCN) that predicts the molecular properties of a molecule using the atom and functional group information as inputs. Molecules can contain many types of functional groups, which will affect the properties the molecules. For example, the toxicity of a molecule is associated with toxicophores, such as nitroaromatic groups and thiourea. Conventional graph-based methods that consider the pair-wise interactions between nodes are inefficient in expressing the complex relationship between multiple nodes in a graph flexibly, and applying multi-hops may result in oversmoothing and overfitting problems. Hence, we propose MolHGCN to capture the substructural difference between molecules using the atom and functional group information. MolHGCN constructs a hypergraph representation of a molecule using functional group information from the input SMILES strings, extracts hidden representation using a two-stage message passing process (atom and functional group message passing), and predicts the properties of the molecules using the extracted hidden representation. We evaluate the performance of our model using Tox21, ClinTox, SIDER, BBBP, BACE, ESOL, FreeSolv and Lipophilicity datasets. We show that our model is able to outperform other baseline methods for most of the datasets. We particularly show that incorporating functional group information along with atom information results in better separability in the latent space, thus increasing the prediction accuracy of the molecule property prediction.

Motivation & Objective

• To address the limitations of conventional graph-based models in capturing complex, multi-node interactions in molecules.
• To improve molecular property prediction by incorporating functional group information alongside atomic structure.
• To reduce oversmoothing and overfitting issues common in multi-hop graph neural networks.
• To enhance feature representation learning through a two-stage message passing process on hypergraphs.
• To demonstrate that functional group-aware representations lead to better separability in the latent space.

Proposed method

• Constructs a hypergraph representation of a molecule using functional group information extracted from SMILES strings.
• Employs a two-stage message passing mechanism: first at the atom level, then at the functional group level.
• Uses hypergraph convolution to aggregate messages across atoms and functional groups simultaneously.
• Integrates both atom and functional group features into a unified hidden representation for property prediction.
• Applies a final readout layer to predict molecular properties using the aggregated representation.
• Trains the model end-to-end using standard backpropagation with mean squared error or binary cross-entropy loss depending on the task.

Experimental results

Research questions

• RQ1 Can hypergraph convolution effectively model multi-node interactions in molecular graphs compared to standard graph convolutions?
• RQ2 Does incorporating functional group information improve the predictive performance of molecular property models?
• RQ3 How does the two-stage message passing mechanism affect representation learning in the latent space?
• RQ4 To what extent does functional group integration enhance feature separability in the learned representations?
• RQ5 Does MolHGCN generalize well across diverse molecular property prediction tasks?

Key findings

• MolHGCN outperforms existing baseline models on most of the evaluated datasets, including Tox21, BACE, and BBBP.
• The integration of functional group information leads to better separability in the latent space, improving model generalization.
• The two-stage message passing process effectively captures substructural differences between molecules.
• The model mitigates oversmoothing and overfitting issues common in multi-hop graph neural networks.
• Functional group-aware representations significantly enhance prediction accuracy, especially for toxicity-related properties.
• The hypergraph-based approach provides a more flexible and expressive way to model complex interactions among atoms and functional groups.