Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Graph Diffusion Network for Drug-Gene Prediction

Jiayang Wu, Wensheng Gan, Philip S. Yu

TL;DR

GDNDGP tackles drug-gene prediction under data sparsity by fusing meta-path-based homogeneous graph learning with a diffusion-driven hard negative sampler for contrastive learning. It introduces meta-paths D-G-D and G-D-G to strengthen intra-type relationships and derives a graph diffusion network to progressively generate challenging negative samples, reducing the need for exhaustive negatives. The method delivers state-of-the-art results on DGIdb 4.0 and strong generalization on drug-go-disease tripartite networks, with ablations showing the diffusion component as a key contributor. The approach offers a scalable, robust framework for heterogeneous biological networks and has potential for extensions to miRNA-drug interactions and other multi-omics tasks.

Abstract

Predicting drug-gene associations is crucial for drug development and disease treatment. While graph neural networks (GNN) have shown effectiveness in this task, they face challenges with data sparsity and efficient contrastive learning implementation. We introduce a graph diffusion network for drug-gene prediction (GDNDGP), a framework that addresses these limitations through two key innovations. First, it employs meta-path-based homogeneous graph learning to capture drug-drug and gene-gene relationships, ensuring similar entities share embedding spaces. Second, it incorporates a parallel diffusion network that generates hard negative samples during training, eliminating the need for exhaustive negative sample retrieval. Our model achieves superior performance on the DGIdb 4.0 dataset and demonstrates strong generalization capability on tripartite drug-gene-disease networks. Results show significant improvements over existing methods in drug-gene prediction tasks, particularly in handling complex heterogeneous relationships. The source code is publicly available at https://github.com/csjywu1/GDNDGP.

Graph Diffusion Network for Drug-Gene Prediction

TL;DR

GDNDGP tackles drug-gene prediction under data sparsity by fusing meta-path-based homogeneous graph learning with a diffusion-driven hard negative sampler for contrastive learning. It introduces meta-paths D-G-D and G-D-G to strengthen intra-type relationships and derives a graph diffusion network to progressively generate challenging negative samples, reducing the need for exhaustive negatives. The method delivers state-of-the-art results on DGIdb 4.0 and strong generalization on drug-go-disease tripartite networks, with ablations showing the diffusion component as a key contributor. The approach offers a scalable, robust framework for heterogeneous biological networks and has potential for extensions to miRNA-drug interactions and other multi-omics tasks.

Abstract

Predicting drug-gene associations is crucial for drug development and disease treatment. While graph neural networks (GNN) have shown effectiveness in this task, they face challenges with data sparsity and efficient contrastive learning implementation. We introduce a graph diffusion network for drug-gene prediction (GDNDGP), a framework that addresses these limitations through two key innovations. First, it employs meta-path-based homogeneous graph learning to capture drug-drug and gene-gene relationships, ensuring similar entities share embedding spaces. Second, it incorporates a parallel diffusion network that generates hard negative samples during training, eliminating the need for exhaustive negative sample retrieval. Our model achieves superior performance on the DGIdb 4.0 dataset and demonstrates strong generalization capability on tripartite drug-gene-disease networks. Results show significant improvements over existing methods in drug-gene prediction tasks, particularly in handling complex heterogeneous relationships. The source code is publicly available at https://github.com/csjywu1/GDNDGP.

Paper Structure

This paper contains 21 sections, 26 equations, 7 figures, 6 tables, 3 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Traditional methods
  4. GNN-based methods
  5. Meta-path based methods
  6. Preliminaries
  7. Algorithm
  8. Meta-path construction
  9. Graph aggregation network
  10. Graph diffusion network
  11. Objective and optimization
  12. Complexity analysis
  13. Experiments
  14. Datasets
  15. Baselines and experimental settings
  16. ...and 6 more sections

Figures (7)

  • Figure 1: An illustration of drug-gene associations (top), meta-path construction (left), and heterogeneous graph message passing (right) used in our framework.
  • Figure 2: An illustration of the graph diffusion network in the GDNDGP. Positive drug-gene pairs are contrasted with two types of negative pairs: simple negative pairs generated from unlinked drug-gene pairs, and hard negative pairs generated through a diffusion process.
  • Figure 3: The ROC-AUC comparisons between two datasets.
  • Figure 4: Performance of GDNDGP across drugs with varying numbers of known gene interactions.
  • Figure 5: Performance comparison with error bars (95% CI).
  • ...and 2 more figures

Theorems & Definitions (5)

  • Definition 1: meta-path
  • Definition 2: homogeneous graph and heterogeneous graph
  • Definition 3: contrastive learning
  • Definition 4: graph convolutional network (GCN)
  • Definition 5: diffusion network