Novel Local and Graph-Based Neural Networks Enhance Link Prediction for Protein-Protein Interactions

Document Type

Conference Proceeding

Source of Publication

Lecture Notes in Networks and Systems

Publication Date

7-2-2026

Abstract

This study investigates the challenging task of link prediction within protein–protein interaction (PPI) networks, targeting the critical issue of incomplete biological knowledge. Notably, only about 20% of yeast proteins and a mere 0.3% of human proteins have been identified to date, highlighting the urgency of this problem. The research primarily focuses on comparing traditional local similarity measures—such as Common Neighbors, Jaccard Coefficient, and Adamic-Adar Index—with more advanced graph-based learning models, including Graph Convolutional Networks (GCNs), Graph Attention Networks (GATs), and GraphSAGE. By conducting a thorough evaluation of these methods, the study offers a comprehensive comparison of their effectiveness and unique properties. The analysis reveals that models based on graph neural networks (GNNs) significantly outperform traditional similarity-based techniques in predicting missing links within PPI networks. These findings highlight the strong potential of GNNs to enhance the predictive accuracy in biological domains, ultimately contributing to a more complete and nuanced understanding of protein interactions and their roles in complex biological systems.

ISBN

[9783032211736]

ISSN

2367-3370

Publisher

Springer Nature Switzerland

Volume

1889 LNNS

First Page

349

Last Page

361

Disciplines

Computer Sciences

Keywords

Complex Network Analysis, Graph Based Method, Link Prediction, Protein-Protein Interaction

Scopus ID

105047674665

Indexed in Scopus

yes

Open Access

no

Link to Full Text

Share

COinS