Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MGKAN: Predicting Asymmetric Drug-Drug Interactions via a Multimodal Graph Kolmogorov-Arnold Network

Kunyi Fan, Mengjie Chen, Longlong Li, Cunquan Qu

TL;DR

The paper addresses the challenge of predicting asymmetric DDIs by introducing MGKAN, a Multimodal Graph Kolmogorov-Arnold Network that replaces MLPs with learnable basis functions $\phi(\cdot)$ in a Graph Kolmogorov-Arnold Network to capture nonlinear, direction-specific interactions. It builds three complementary views—an asymmetric DDI network, a co-interaction network, and a biochemical similarity network—and fuses their embeddings with both linear attention and nonlinear KAN-based mechanisms, culminating in a bilinear decoder for prediction. Empirical results on two DrugBank datasets show MGKAN outperforms seven state-of-the-art baselines across multiple metrics, with ablations confirming the critical roles of KAN, fusion modules, and the three views. The approach advances safe polypharmacy decision-making by more accurately modeling directional drug effects and enabling potential discovery of novel DDIs.

Abstract

Predicting drug-drug interactions (DDIs) is essential for safe pharmacological treatments. Previous graph neural network (GNN) models leverage molecular structures and interaction networks but mostly rely on linear aggregation and symmetric assumptions, limiting their ability to capture nonlinear and heterogeneous patterns. We propose MGKAN, a Graph Kolmogorov-Arnold Network that introduces learnable basis functions into asymmetric DDI prediction. MGKAN replaces conventional MLP transformations with KAN-driven basis functions, enabling more expressive and nonlinear modeling of drug relationships. To capture pharmacological dependencies, MGKAN integrates three network views-an asymmetric DDI network, a co-interaction network, and a biochemical similarity network-with role-specific embeddings to preserve directional semantics. A fusion module combines linear attention and nonlinear transformation to enhance representational capacity. On two benchmark datasets, MGKAN outperforms seven state-of-the-art baselines. Ablation studies and case studies confirm its predictive accuracy and effectiveness in modeling directional drug effects.

MGKAN: Predicting Asymmetric Drug-Drug Interactions via a Multimodal Graph Kolmogorov-Arnold Network

TL;DR

The paper addresses the challenge of predicting asymmetric DDIs by introducing MGKAN, a Multimodal Graph Kolmogorov-Arnold Network that replaces MLPs with learnable basis functions in a Graph Kolmogorov-Arnold Network to capture nonlinear, direction-specific interactions. It builds three complementary views—an asymmetric DDI network, a co-interaction network, and a biochemical similarity network—and fuses their embeddings with both linear attention and nonlinear KAN-based mechanisms, culminating in a bilinear decoder for prediction. Empirical results on two DrugBank datasets show MGKAN outperforms seven state-of-the-art baselines across multiple metrics, with ablations confirming the critical roles of KAN, fusion modules, and the three views. The approach advances safe polypharmacy decision-making by more accurately modeling directional drug effects and enabling potential discovery of novel DDIs.

Abstract

Predicting drug-drug interactions (DDIs) is essential for safe pharmacological treatments. Previous graph neural network (GNN) models leverage molecular structures and interaction networks but mostly rely on linear aggregation and symmetric assumptions, limiting their ability to capture nonlinear and heterogeneous patterns. We propose MGKAN, a Graph Kolmogorov-Arnold Network that introduces learnable basis functions into asymmetric DDI prediction. MGKAN replaces conventional MLP transformations with KAN-driven basis functions, enabling more expressive and nonlinear modeling of drug relationships. To capture pharmacological dependencies, MGKAN integrates three network views-an asymmetric DDI network, a co-interaction network, and a biochemical similarity network-with role-specific embeddings to preserve directional semantics. A fusion module combines linear attention and nonlinear transformation to enhance representational capacity. On two benchmark datasets, MGKAN outperforms seven state-of-the-art baselines. Ablation studies and case studies confirm its predictive accuracy and effectiveness in modeling directional drug effects.
Paper Structure (11 sections, 15 equations, 3 figures, 2 tables)

This paper contains 11 sections, 15 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Methodology
  4. Feature Extraction Module
  5. Multimodal Fusion Module
  6. Asymmetric DDI Prediction and Loss Function
  7. Experiments
  8. Performance Comparison
  9. Ablation Study
  10. Case Study
  11. Conclusion

Figures (3)

  • Figure 1: Examples of symmetric and asymmetric DDIs.
  • Figure 2: The overall framework of MGKAN.
  • Figure 3: Ablation study results of MGKAN and its variants. (A) Performance on Task 1. (B) Performance on Task 2.