Multimodal Contrastive Representation Learning in Augmented Biomedical Knowledge Graphs

2025-01-03

AI Summary

Link prediction in biomedical knowledge graphs (BKGs) remains challenging due to sparse, heterogeneous entity representations and limited generalization to unseen nodes. Method: This paper proposes a multimodal collaborative modeling framework that jointly integrates language model–derived semantic representations with graph contrastive learning. It unifies multimodal contrastive learning with knowledge graph embedding, jointly encoding intra-entity multimodal features (e.g., biological sequences and textual descriptions) and inter-entity topological structures. We further construct PrimeKG++, an enhanced BKG explicitly designed to support zero-shot node link prediction. Contribution/Results: Our approach achieves significant improvements in link prediction performance on both PrimeKG++ and DrugBank benchmarks, demonstrating strong cross-task and cross-domain generalization. All code, pretrained models, and datasets are publicly released to foster reproducibility and community advancement.

Technology Category

• Data Mining & Knowledge Management: Linked Open Data, Knowledge Graphs & KB Completion
• Machine Learning: Multimodal Learning
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG

Abstract

Biomedical Knowledge Graphs (BKGs) integrate diverse datasets to elucidate complex relationships within the biomedical field. Effective link prediction on these graphs can uncover valuable connections, such as potential novel drug-disease relations. We introduce a novel multimodal approach that unifies embeddings from specialized Language Models (LMs) with Graph Contrastive Learning (GCL) to enhance intra-entity relationships while employing a Knowledge Graph Embedding (KGE) model to capture inter-entity relationships for effective link prediction. To address limitations in existing BKGs, we present PrimeKG++, an enriched knowledge graph incorporating multimodal data, including biological sequences and textual descriptions for each entity type. By combining semantic and relational information in a unified representation, our approach demonstrates strong generalizability, enabling accurate link predictions even for unseen nodes. Experimental results on PrimeKG++ and the DrugBank drug-target interaction dataset demonstrate the effectiveness and robustness of our method across diverse biomedical datasets. Our source code, pre-trained models, and data are publicly available at https://github.com/HySonLab/BioMedKG

Problem

• Biomedical Knowledge Graphs
• Text and Graph Integration
• Drug-Disease Prediction

Innovation

• Integrated Text and Graph Information
• Predictive Modeling for Drug-Disease Associations
• Comprehensive Biomedical Knowledge Graph PrimeKG++