Research map / Graph learning / Knowledge and temporal graphs
Biomedical Knowledge Graph Embedding with Householder Projection (Student Abstract)
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes CDHse, a biomedical knowledge graph embedding model for drug-drug interaction prediction that fuses PubMedBERT/CNN text features and GNN molecular features and applies Householder projections in complex space.
Problem
Drug-drug interaction prediction on biomedical knowledge graphs ignores molecular structure and position features and struggles to model 1-N, N-1, N-N relational mapping properties.
Approach
Combines BERT-based position features (CNN) and GNN molecular-structure features of drug pairs into a synthetic entity feature, then applies relation-specific Householder projections with complex-space embedding (rotation of entities) to model relational mapping properties.
Key results
- On OGB-Biokg, CDHse reaches F1 91.27 versus 90.64 for KG2ECapsule.
- On DrugBank, CDHse reaches F1 92.27 versus 90.64 for KG2ECapsule.
- On KEGG, CDHse reaches F1 53.24 versus 51.31 for KG2ECapsule.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Replace KG2ECapsule with CDHse | Drug-drug interaction prediction · DrugBank | F1: 90.64 -> 92.27 | Helped |
| Replace KG2ECapsule with CDHse | Drug-drug interaction prediction · KEGG | F1: 51.31 -> 53.24 | Helped |
| Replace KG2ECapsule with CDHse | Drug-drug interaction prediction · OGB-Biokg | F1: 90.64 -> 91.27 | Helped |
Limitations
- Evaluated only on three BioKG benchmarks with a short student abstract format
- Householder projection details deferred to an appendix not included in the text
- Comparison limited to a small set of KG-based baselines
Experimental setup
Datasets: OGB-Biokg DrugBank KEGG
Benchmarks: OGB-Biokg DrugBank KEGG
Metrics: Precision Recall F1
Baselines: BERTKG-DDI Xin et al. (Jin and Sun 2022) KG2ECapsule
Builds on: KG2ECapsule RotatE
Closest papers by meaning
- Dual-Channel Learning Framework for Drug-Drug Interaction Prediction via Relation-Aware Heterogeneous Graph Transformer (AAAI 2024 · similarity 0.88)
- Knowledge Enhanced Representation Learning for Drug Discovery (AAAI 2024 · similarity 0.87)
- Bi-level Contrastive Learning for Knowledge-Enhanced Molecule Representations (AAAI 2025 · similarity 0.85)
- MKG-FENN: A Multimodal Knowledge Graph Fused End-to-End Neural Network for Accurate Drug–Drug Interaction Prediction (AAAI 2024 · similarity 0.84)
- On the Scalability of GNNs for Molecular Graphs (NeurIPS 2024 · similarity 0.84)
- Pre-Training Graph Neural Networks on Molecules by Using Subgraph-Conditioned Graph Information Bottleneck (AAAI 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.