{
  "id": 19426405,
  "title": "HV-MultiScaleGNN: A Multi-Scale Graph Neural Network Framework for Host-Virus Protein-Protein Interactions Prediction",
  "url": "https://urgent.news/2026/11/10/hv-multiscalegnn-a-multi-scale-graph-neural-network-framework-for",
  "topic": "ai",
  "section": "AI",
  "published": "2026-11-10T00:00:00.000Z",
  "source": {
    "name": "bioRxiv",
    "slug": "biorxiv",
    "url": "https://www.biorxiv.org/content/10.64898/2026.11.08.765214v1?rss=1"
  },
  "original_language": "en",
  "account": "Accurate identification of human-virus protein-protein interactions (HV-PPIs) is crucial to understanding viral infection mechanisms and advancing antiviral therapeutic discovery. Current experimental methods, however, are time-consuming and costly, and many computational approaches rely on incomplete protein sequence, structural, and interaction data. To address these challenges, researchers have developed HV-MultiScaleGNN, a topology-driven multi-scale graph neural network (GNN) framework specifically designed for HV-PPIs prediction.\n\nHV-MultiScaleGNN exclusively utilizes the topology of a heterogeneous interaction network, as opposed to relying on sequence or structural information. The framework approaches HV-PPIs prediction as a heterogeneous link prediction problem by combining the human interactome with experimentally validated HV-PPIs. It systematically explores three complementary graph representation strategies: Local, Global, and an integrated Local+Global view.\n\nThe Local view employs one-hop relation-aware graph attention to model fine-grained neighborhood interactions. Meanwhile, the Global view captures higher-order structural dependencies through propagation-based neighborhood aggregation. The integrated Local+Global view combines both strategies, seamlessly encoding local connectivity and global topological context within a unified embedding space. This enables relation-specific link prediction without relying on sequence- or structure-derived features.\n\nTo evaluate the effectiveness of the three strategies, HV-MultiScaleGNN employs discrimination, calibration, and ranking metrics on large-scale HV-PPIs datasets. The Local view emerges as the best discriminator, achieving an Area Under the ROC Curve (AUROC) of 0.9444 and an Area Under the Precision-Recall Curve (AUPRC) of 0.9422. The Global view, on the other hand, demonstrates the lowest expected calibration error (ECE) of 0.0379.\n\nThe integrated Local+Global view consistently delivers strong predictive performance, with an AUROC of 0.9378 and an AUPRC of 0.9356. Comparative evaluation against representative graph representation learning methods reveals HV-MultiScaleGNN's competitive performance. The framework also identifies interfacial interactions through protein-protein docking and explores atomistic-level understanding. Furthermore, functional insights into selected host proteins are gained through Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis.\n\nBy systematically evaluating complementary graph representation strategies for heterogeneous biological link prediction, HV-MultiScaleGNN paves the way for experimental researchers to discover novel HV-PPIs.",
  "summary": "Accurate identification of HV-PPIs is essential for understanding viral infection mechanisms and for developing antiviral therapeutic discovery. However, the experimental identification process is often time-consuming and expensive, while many computational approaches depend on incomplete protein sequences, structural and interaction information. Here, we present HV-MultiScaleGNN, a…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}