Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ResMGCN: Residual Message Graph Convolution Network for Fast Biomedical Interactions Discovering

Zecheng Yin

TL;DR

This work introduces ResMGCN, a residual-message graph convolution network designed for fast and accurate biomedical interaction prediction. By fusing the current-layer messages with residual information from the previous layer through a fusion function, ResMGCN preserves low-order and high-order information without heavy memory or computation costs. Evaluation on four public biomedical interaction networks (DTI, DDI, PPI, GDI) shows that ResMGCN achieves state-of-the-art or competitive predictive accuracy while substantially reducing training time compared to prior multi-hop methods. The approach offers scalable, end-to-end link prediction on complex biomedical graphs and holds promise for broader graph-based biomedical analytics.

Abstract

Biomedical information graphs are crucial for interaction discovering of biomedical information in modern age, such as identification of multifarious molecular interactions and drug discovery, which attracts increasing interests in biomedicine, bioinformatics, and human healthcare communities. Nowadays, more and more graph neural networks have been proposed to learn the entities of biomedical information and precisely reveal biomedical molecule interactions with state-of-the-art results. These methods remedy the fading of features from a far distance but suffer from remedying such problem at the expensive cost of redundant memory and time. In our paper, we propose a novel Residual Message Graph Convolution Network (ResMGCN) for fast and precise biomedical interaction prediction in a different idea. Specifically, instead of enhancing the message from far nodes, ResMGCN aggregates lower-order information with the next round higher information to guide the node update to obtain a more meaningful node representation. ResMGCN is able to perceive and preserve various messages from the previous layer and high-order information in the current layer with least memory and time cost to obtain informative representations of biomedical entities. We conduct experiments on four biomedical interaction network datasets, including protein-protein, drug-drug, drug-target, and gene-disease interactions, which demonstrates that ResMGCN outperforms previous state-of-the-art models while achieving superb effectiveness on both storage and time.

ResMGCN: Residual Message Graph Convolution Network for Fast Biomedical Interactions Discovering

TL;DR

This work introduces ResMGCN, a residual-message graph convolution network designed for fast and accurate biomedical interaction prediction. By fusing the current-layer messages with residual information from the previous layer through a fusion function, ResMGCN preserves low-order and high-order information without heavy memory or computation costs. Evaluation on four public biomedical interaction networks (DTI, DDI, PPI, GDI) shows that ResMGCN achieves state-of-the-art or competitive predictive accuracy while substantially reducing training time compared to prior multi-hop methods. The approach offers scalable, end-to-end link prediction on complex biomedical graphs and holds promise for broader graph-based biomedical analytics.

Abstract

Biomedical information graphs are crucial for interaction discovering of biomedical information in modern age, such as identification of multifarious molecular interactions and drug discovery, which attracts increasing interests in biomedicine, bioinformatics, and human healthcare communities. Nowadays, more and more graph neural networks have been proposed to learn the entities of biomedical information and precisely reveal biomedical molecule interactions with state-of-the-art results. These methods remedy the fading of features from a far distance but suffer from remedying such problem at the expensive cost of redundant memory and time. In our paper, we propose a novel Residual Message Graph Convolution Network (ResMGCN) for fast and precise biomedical interaction prediction in a different idea. Specifically, instead of enhancing the message from far nodes, ResMGCN aggregates lower-order information with the next round higher information to guide the node update to obtain a more meaningful node representation. ResMGCN is able to perceive and preserve various messages from the previous layer and high-order information in the current layer with least memory and time cost to obtain informative representations of biomedical entities. We conduct experiments on four biomedical interaction network datasets, including protein-protein, drug-drug, drug-target, and gene-disease interactions, which demonstrates that ResMGCN outperforms previous state-of-the-art models while achieving superb effectiveness on both storage and time.
Paper Structure (21 sections, 25 equations, 4 figures, 3 tables)

This paper contains 21 sections, 25 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Preliminaries
  4. Message Passing Framework
  5. Graph Convolution(GCN)
  6. Proposed Residual GCN
  7. Message Passing in GCN
  8. Residual Message GCN (ResMGCN)
  9. ResMGCN for biomedical interaction prediction
  10. Graph ResMGCN Feature Encoder
  11. Graph Predictor
  12. Model Training and Optimization
  13. Experiment
  14. Datasets
  15. Baselines
  16. ...and 6 more sections

Figures (4)

  • Figure 1: Message passing framework illustration. Message step takes representations of multiple nearby neighbours to calculate the message the node needed for updating, formulated in Eq.\ref{['eq:message']}. Passing step updates the node representation according to the message provided by the first step and its own original representation, formulated in Eq.\ref{['eq:passing']}.
  • Figure 2: Illustration of multiplayer of message passing. Node $v_i$ is able to capture messages from node $v_j$ via node $v_k$ by stacking two layers of graph convolution.
  • Figure 3: Illustration of message passing. The layer outputs the updated node representation $X$ for the next layer. The messages are left within the current layer.
  • Figure 4: Illustration of ResMGCN. Besides of the updated representations, layer keeps the messages of current layer and pass the hierarchy information for next layer. Each layer has a information fusion function $\Phi$ to manage the information received for calculating the guided information $C$ for node updating.