GSNMF: A Graph-Regularized Nonnegative Matrix Factorization Framework With Biological Priors for Protein-Protein Interaction Prediction

Accurate prediction of protein-protein interactions (PPIs) supports the reconstruction of cellular networks and the prioritization of experimentally testable hypotheses. Network-based prediction is challenged by sparse, noisy, and incomplete interaction data, motivating the integration of complementary biological information. We propose GSNMF, a graph-regularized nonnegative matrix factorization framework that combines PPI topology and Gene Ontology (GO) semantic similarity through a shared latent basis. View-specific coefficient matrices reconstruct the topological and functional information, while graph Laplacian regularization encourages neighborhood consistency and an element-wise $L_{p}$ penalty, with $1< p< 2$, regularizes the shared basis. The reconstructed interaction scores are symmetrized and subsequently adjusted using subcellular-localization compatibility. Soft localization weighting attenuates scores for protein pairs without shared annotations rather than excluding them outright. Evaluation on five Saccharomyces cerevisiae PPI datasets shows competitive performance against representative matrix-factorization baselines. Localization ablations illustrate the tradeoff between prioritizing compatible pairs and suppressing recorded interactions lacking shared localization annotations. On collins2007 and DIP core, soft weighting preserves nearly the same P@100 as hard filtering while yielding higher AUPRC. These findings support the use of complementary functional information and localization-aware candidate prioritization within a shared-factorization framework. Because unobserved pairs are provisionally treated as negatives, the reported performance measures recovery of recorded interactions rather than definitive biological interaction accuracy.

Authors

Institutions

Publication Details

Journal
IEEE Transactions on Computational Biology and Bioinformatics
Published
2026-11-20
DOI
https://doi.org/10.1109/tcbbio.2026.3756406
Primary Topic
Bioinformatics and Genomic Networks
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

GSNMF: A Graph-Regularized Nonnegative Matrix Factorization Framework With Biological Priors for Protein-Protein Interaction Prediction

Yan Yan, Tianhai Tian, Wei Zhang
IEEE Transactions on Computational Biology and Bioinformatics
Bioinformatics and Genomic Networks
article

GSNMF: A Graph-Regularized Nonnegative Matrix Factorization Framework With Biological Priors for Protein-Protein Interaction Prediction

Yan Yan, Tianhai Tian, Wei Zhang
article en

Abstract

Accurate prediction of protein-protein interactions (PPIs) supports the reconstruction of cellular networks and the prioritization of experimentally testable hypotheses. Network-based prediction is challenged by sparse, noisy, and incomplete interaction data, motivating the integration of complementary biological information. We propose GSNMF, a graph-regularized nonnegative matrix factorization framework that combines PPI topology and Gene Ontology (GO) semantic similarity through a shared latent basis. View-specific coefficient matrices reconstruct the topological and functional information, while graph Laplacian regularization encourages neighborhood consistency and an element-wise $L_{p}$ penalty, with $1< p< 2$, regularizes the shared basis. The reconstructed interaction scores are symmetrized and subsequently adjusted using subcellular-localization compatibility. Soft localization weighting attenuates scores for protein pairs without shared annotations rather than excluding them outright. Evaluation on five Saccharomyces cerevisiae PPI datasets shows competitive performance against representative matrix-factorization baselines. Localization ablations illustrate the tradeoff between prioritizing compatible pairs and suppressing recorded interactions lacking shared localization annotations. On collins2007 and DIP core, soft weighting preserves nearly the same P@100 as hard filtering while yielding higher AUPRC. These findings support the use of complementary functional information and localization-aware candidate prioritization within a shared-factorization framework. Because unobserved pairs are provisionally treated as negatives, the reported performance measures recovery of recorded interactions rather than definitive biological interaction accuracy.

IEEE Transactions on Computational Biology and BioinformaticsVol. PP
Monash University (AU), Wuhan Institute of Technology (CN)
Bioinformatics and Genomic Networks
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon
GSNMF: A Graph-Regularized Nonnegative Matrix Factorization Framework With Biological Priors for Protein-Protein Interaction Prediction — Yan Yan, Tianhai Tian, et al. · IEEE Transactions on Computational Biology and Bioinformatics (2026) | TGRS Research Map | TGRS