GeoScatt-GNN: A Geometric Scattering Transform-Based Graph Neural Network Model for Ames Mutagenicity Prediction

TL;DR

A hybrid approach that combines geometric graph scattering, Graph Isomorphism Networks, and machine learning models, achieving strong results in mutagenicity prediction is presented, and a novel graph neural network architecture, MOLG3-SAGE, which integrates GGS node features into a fully connected graph structure is introduced.

Abstract

This paper tackles the pressing challenge of mutagenicity prediction by introducing three ground-breaking approaches. First, it showcases the superior performance of 2D scattering coefficients extracted from molecular images, compared to traditional molecular descriptors. Second, it presents a hybrid approach that combines geometric graph scattering (GGS), Graph Isomorphism Networks (GIN), and machine learning models, achieving strong results in mutagenicity prediction. Third, it introduces a novel graph neural network architecture, MOLG3-SAGE, which integrates GGS node features into a fully connected graph structure, delivering outstanding predictive accuracy. Experimental results on the ZINC dataset demonstrate significant improvements, emphasizing the effectiveness of blending 2D and geometric scattering techniques with graph neural networks. This study illustrates the potential of GNNs and GGS for mutagenicity prediction, with broad implications for drug discovery and chemical safety assessment.

Figures (6)

• Figure 1: 2D Morlet Wavelet visualization at various orientations and parameters
• Figure 2: Graph Scattering Transform with $J=3$ and $L=3$ for multiscale signal decomposition.
• Figure 3: Multi-Modal Pipeline for Mutagenicity Prediction Using Molecule Representations as Graphs and Images. The pipeline utilizes the Hansen et al. dataset of 6,512 compounds divided into mutagens and non-mutagens. Molecules are transformed into graph representations for geometric scattering (a) using Diffusion and Tight HANN wavelets, along with embeddings from a Graph Isomorphism Network (GIN), and into 2D molecular images for scattering coefficient extraction (b). These features are fused and fed into machine learning models to classify compounds as mutagenic or non-mutagenic.
• Figure 4: Fully Connected Layer Architecture for Binary Mutagenicity Classification in MolG³-SAGE Framework
• Figure 5: Comparative Performance Analysis of Machine Learning Models for Ames Mutagenecity Using Multiple Evaluation Metrics