Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A deep graph model for the signed interaction prediction in biological network

Shuyi Jin, Mengji Zhang, Meijie Wang, Lun Yu

TL;DR

This work introduces RGCNTD, a polarity-aware deep graph model that jointly predicts polar (increase/decrease) and non-polar (binding/affect) chemical-gene interactions in a heterogeneous network. By coupling a two-layer GCN encoder with an RGCN-based tensor decomposition decoder and a conflict-aware CL sampling strategy, the model achieves superior discrimination of polarity and overall predictive accuracy, outperforming multiple baselines. The authors also propose novel metrics, $AUC_{polarity}$ and CP@500, to specifically evaluate polarity differentiation and high-confidence predictions, and show that, in some cases, additional subgraph inputs can harm performance. The framework advances network pharmacology by enabling robust, polarity-aware predictions on large Chemical-Gene interaction networks, with practical implications for drug repurposing and mechanism interpretation.

Abstract

Predicting signed interactions in biological networks is crucial for understanding drug mechanisms and facilitating drug repurposing. While deep graph models have demonstrated success in modeling complex biological systems, existing approaches often fail to distinguish between positive and negative interactions, limiting their utility for precise pharmacological predictions. In this study, we propose a novel deep graph model, \textbf{RGCNTD} (Relational Graph Convolutional Network with Tensor Decomposition), designed to predict both polar (e.g., activation, inhibition) and non-polar (e.g., binding, affect) chemical-gene interactions. Our model integrates graph convolutional networks with tensor decomposition to enhance feature representation and incorporates a conflict-aware sampling strategy to resolve polarity ambiguities. We introduce new evaluation metrics, \textit{AUC\textsubscript{polarity}} and \textit{CP@500}, to assess the model's ability to differentiate interaction types. Experimental results demonstrate that \textbf{RGCNTD} outperforms baseline models, achieving superior classification accuracy and improved discrimination of polar edges. Furthermore, we analyze the impact of subgraph components on predictive performance, revealing that additional network structures do not always enhance accuracy. These findings highlight the importance of polarity-aware modeling in drug discovery and network pharmacology, providing a robust framework for predicting complex biological interactions.

A deep graph model for the signed interaction prediction in biological network

TL;DR

This work introduces RGCNTD, a polarity-aware deep graph model that jointly predicts polar (increase/decrease) and non-polar (binding/affect) chemical-gene interactions in a heterogeneous network. By coupling a two-layer GCN encoder with an RGCN-based tensor decomposition decoder and a conflict-aware CL sampling strategy, the model achieves superior discrimination of polarity and overall predictive accuracy, outperforming multiple baselines. The authors also propose novel metrics, and CP@500, to specifically evaluate polarity differentiation and high-confidence predictions, and show that, in some cases, additional subgraph inputs can harm performance. The framework advances network pharmacology by enabling robust, polarity-aware predictions on large Chemical-Gene interaction networks, with practical implications for drug repurposing and mechanism interpretation.

Abstract

Predicting signed interactions in biological networks is crucial for understanding drug mechanisms and facilitating drug repurposing. While deep graph models have demonstrated success in modeling complex biological systems, existing approaches often fail to distinguish between positive and negative interactions, limiting their utility for precise pharmacological predictions. In this study, we propose a novel deep graph model, \textbf{RGCNTD} (Relational Graph Convolutional Network with Tensor Decomposition), designed to predict both polar (e.g., activation, inhibition) and non-polar (e.g., binding, affect) chemical-gene interactions. Our model integrates graph convolutional networks with tensor decomposition to enhance feature representation and incorporates a conflict-aware sampling strategy to resolve polarity ambiguities. We introduce new evaluation metrics, \textit{AUC\textsubscript{polarity}} and \textit{CP@500}, to assess the model's ability to differentiate interaction types. Experimental results demonstrate that \textbf{RGCNTD} outperforms baseline models, achieving superior classification accuracy and improved discrimination of polar edges. Furthermore, we analyze the impact of subgraph components on predictive performance, revealing that additional network structures do not always enhance accuracy. These findings highlight the importance of polarity-aware modeling in drug discovery and network pharmacology, providing a robust framework for predicting complex biological interactions.
Paper Structure (22 sections, 11 equations, 4 figures, 7 tables)

This paper contains 22 sections, 11 equations, 4 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Results
  3. Performance Of Models Prediction
  4. Conflicting Edge Sampling Strategy
  5. Designed Metrics for Polairty Prediction
  6. Impact of Network Component
  7. Method
  8. Model Construction
  9. Network Construction
  10. Handling Polar Edges in Chemical-Gene Prediction Tasks
  11. Baseline
  12. Experimental Setup
  13. Metrics Design
  14. Average Precision at Polar Edges
  15. Polarity Degree (C)-based Precision
  16. ...and 7 more sections

Figures (4)

  • Figure 1: To model a triplet pair within the network, the edges could be polar (e.g., positive or negative) or non-polar (e.g., neutral). In cases where polarity is present, the model is expected to accurately distinguish and identify the edge’s polarity.
  • Figure 2: The figures illustrate the performance of four models (RGCNTD, RGCN, GraphSAGE, and TransE) in polarity edge analysis. The left-side plots show the models’ performance without the CL sampling strategy, while the right-side plots display results after incorporating the strategy. The log$_2$-transformed ranking lines connecting the two sides highlight each model’s ability to differentiate between the two types of interactions
  • Figure 3: Overview of model architecture
  • Figure 4: The distribution of $C$ in the CL model and NO CL model before and after the mathematical transformation.