MKGR: Multimodal Knowledge-Graph Representation Learning for Cold-Start Protein-Protein Interaction Prediction

2026-07-02 · Source: Machine Learning · Field: Science & Research — Life Sciences & Biology, Health & Medical Research, Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

Accurate protein-protein interaction (PPI) prediction is central to functional genomics, disease mechanism discovery, and drug development. A difficult setting arises when candidate interactions include proteins that have no observed PPI edges during training, where models relying on network topology alone often lose useful context. The MKGR framework addresses this by combining region-aware protein sequence encoding with four protein-centered biomedical knowledge graphs: protein-drug, protein-disease, protein-miRNA, and protein-lncRNA associations. It employs a sequence branch for contextual representations and graph attention encoders for modality-specific protein embeddings. A bridge reconstruction objective regularizes graph learning, and a pair-level gating module integrates sequence and graph evidence. Experiments on two benchmark datasets demonstrate MKGR's consistent outperformance of competitive baselines across ACC, F1, AUC, AUPR, and MCC in novel-old and novel-novel cold-start settings.

Key takeaway

For AI Scientists and Research Scientists focused on advancing protein-protein interaction prediction, particularly in cold-start scenarios involving novel proteins, the MKGR framework presents a significant methodological advancement. You should consider adopting multimodal representation learning, integrating diverse biomedical knowledge graphs alongside protein sequence data, to enhance predictive accuracy and robustness for unobserved interactions.

Key insights

MKGR combines protein sequence and knowledge graph data for robust cold-start protein-protein interaction prediction.

Principles

• Multimodal data improves cold-start PPI prediction.
• Region-aware sequence encoding captures structural context.
• Graph attention encoders learn modality-specific embeddings.

Method

MKGR uses sequence encoding and graph attention encoders on four biomedical KGs, regularized by a bridge reconstruction objective, then integrates evidence via a pair-level gating module.

In practice

• Integrate diverse biomedical KGs for protein analysis.
• Use region-aware sequence features in deep learning.
• Apply gating modules for multimodal evidence fusion.

Topics

• Protein-Protein Interaction
• Knowledge Graphs
• Multimodal Learning
• Cold-Start Prediction
• Protein Sequence Encoding
• Graph Attention Networks

Best for: AI Scientist, Research Scientist

Related on AIssential

• Multi-to-uni modal knowledge transfer pre-training for molecular representation learning — M2UMol enables robust molecular representation learning from incomplete multimodal data by transferring knowledge to a 2D encoder.
• Viral Proteins Reveal Geometry of Protein Language Models — Protein language models organize sequence embeddings by a "nativeness axis" while preserving specific biological group signals.
• EssentialGIN: a new approach for gene essentiality prediction based on graph isomorphism neural networks — Graph isomorphism networks, by integrating biological data and preserving network topology, significantly improve essential gene prediction accuracy.
• Two-dimensional geometric template diffusion for boosting single-sequence protein structure prediction — TDFold uses 2D geometric template diffusion for efficient, accurate single-sequence protein structure prediction.
• 🔬ESM: The Bitter Lesson is Coming for Proteins - Alex Rives, BioHub — Scaling protein language models with vast, diverse data reveals emergent biological capabilities and enables programmable design.
• Assessing uncertainty of sequence representations generated by protein language models — A new model-agnostic measure quantifies the reliability of protein language model sequence representations.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.