Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Learning to Denoise Biomedical Knowledge Graph for Robust Molecular Interaction Prediction

Tengfei Ma, Yujie Chen, Wen Tao, Dashun Zheng, Xuan Lin, Patrick Cheong-lao Pang, Yiping Liu, Yijun Wang, Longyue Wang, Bosheng Song, Xiangxiang Zeng, Philip S. Yu

TL;DR

Experimental results on real-world datasets show that BioKDN surpasses state-of-the-art models in DTI and DDI prediction tasks, confirming the effectiveness and robustness of BioKDN in denoising unreliable interactions within contaminated KGs.

Abstract

Molecular interaction prediction plays a crucial role in forecasting unknown interactions between molecules, such as drug-target interaction (DTI) and drug-drug interaction (DDI), which are essential in the field of drug discovery and therapeutics. Although previous prediction methods have yielded promising results by leveraging the rich semantics and topological structure of biomedical knowledge graphs (KGs), they have primarily focused on enhancing predictive performance without addressing the presence of inevitable noise and inconsistent semantics. This limitation has hindered the advancement of KG-based prediction methods. To address this limitation, we propose BioKDN (Biomedical Knowledge Graph Denoising Network) for robust molecular interaction prediction. BioKDN refines the reliable structure of local subgraphs by denoising noisy links in a learnable manner, providing a general module for extracting task-relevant interactions. To enhance the reliability of the refined structure, BioKDN maintains consistent and robust semantics by smoothing relations around the target interaction. By maximizing the mutual information between reliable structure and smoothed relations, BioKDN emphasizes informative semantics to enable precise predictions. Experimental results on real-world datasets show that BioKDN surpasses state-of-the-art models in DTI and DDI prediction tasks, confirming the effectiveness and robustness of BioKDN in denoising unreliable interactions within contaminated KGs

Learning to Denoise Biomedical Knowledge Graph for Robust Molecular Interaction Prediction

TL;DR

Experimental results on real-world datasets show that BioKDN surpasses state-of-the-art models in DTI and DDI prediction tasks, confirming the effectiveness and robustness of BioKDN in denoising unreliable interactions within contaminated KGs.

Abstract

Molecular interaction prediction plays a crucial role in forecasting unknown interactions between molecules, such as drug-target interaction (DTI) and drug-drug interaction (DDI), which are essential in the field of drug discovery and therapeutics. Although previous prediction methods have yielded promising results by leveraging the rich semantics and topological structure of biomedical knowledge graphs (KGs), they have primarily focused on enhancing predictive performance without addressing the presence of inevitable noise and inconsistent semantics. This limitation has hindered the advancement of KG-based prediction methods. To address this limitation, we propose BioKDN (Biomedical Knowledge Graph Denoising Network) for robust molecular interaction prediction. BioKDN refines the reliable structure of local subgraphs by denoising noisy links in a learnable manner, providing a general module for extracting task-relevant interactions. To enhance the reliability of the refined structure, BioKDN maintains consistent and robust semantics by smoothing relations around the target interaction. By maximizing the mutual information between reliable structure and smoothed relations, BioKDN emphasizes informative semantics to enable precise predictions. Experimental results on real-world datasets show that BioKDN surpasses state-of-the-art models in DTI and DDI prediction tasks, confirming the effectiveness and robustness of BioKDN in denoising unreliable interactions within contaminated KGs
Paper Structure (44 sections, 11 equations, 9 figures, 7 tables)

This paper contains 44 sections, 11 equations, 9 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. KG-based Molecular Interaction Prediction
  4. Denoising Methods on Graphs
  5. Methodology
  6. Preliminaries
  7. Biomedical Knowledge Graph
  8. Local Subgraph
  9. Semantic Subgraph
  10. Problem Definition
  11. Initialization of Knowledge Graph
  12. Subgraph Denoising
  13. Structure Reliability Learning
  14. Smooth Semantic Preservation
  15. MI Maximization in Subgraphs
  16. ...and 29 more sections

Figures (9)

  • Figure 1: The explanatory case of noise within DRKG. Chebi:28300 and DB00130 indicate the same drug L-Glutamine, but they are treated as different entities, which results in entity unalignment and facts missing. Meanwhile, the source document represents the gene MSLN as a biomarker for cancer patients, and no confidence indicates it has a role in the disease Mental Disorders, which introduces noisy interactions into the KG.
  • Figure 2: An example of extracting local (structural reliability) and semantic subgraphs (semantic consistency) surrounding the target molecular pair $(u,v)$.
  • Figure 3: The BioKDN framework comprises three modules for predicting links in a given KG: (a) Initializing the entity and relation embeddings of the KG using RotatE; (b) Denoising the noisy interactions around the predicted link by learning reliable structure and preserving consistent semantics; (c) Maximizing the mutual information between refined and semantic subgraphs to focus on informative interactions.
  • Figure 4: The relation distribution of smoothed KG.
  • Figure 5: Hyper-parameter sensitivity analysis of DTI prediction based on the DrugBank dataset.
  • ...and 4 more figures