PRInter: an inductive multimodal heterogeneous network embedding framework with contrastive learning for lncRNA-protein interaction prediction
- 1Department of Biomedical Engineering, School of Control Science and Engineering, Shandong University, Jinan, 250061 Shandong China
- 2School of Bioengineering, Dalian University of Technology, Dalian, 116024 Liaoning China
- 3National Center for Applied Mathematics, Shandong University, Jinan, 250100 Shandong China
Abstract
Background: Long non-coding RNAs (lncRNAs) regulate diverse biological processes via protein interactions, and disruptions in these interactions are frequently associated with complex diseases. Network-based computational methods show promise in predicting such interactions, yet most cannot generalize to new, unseen molecules.
Results: Here, we introduce PRInter, an inductive multimodal framework integrating sequence features with heterogeneous network structures for lncRNA-protein interaction prediction. PRInter employs meta-path-guided random walks to capture network context and utilizes contrastive learning to refine molecular sequence embeddings, facilitating the prediction of candidate lncRNA-protein pairs involving molecules beyond the training network. To improve the reliability of evaluation, we adopt a degree-balanced negative sampling strategy to reduce potential bias from scale-free network topology. Comprehensive experiments on three public datasets show that PRInter achieves an AUC of 0.882 and performs favorably compared with representative baseline methods.
Conclusions: As an inductive and less biased model, PRInter offers a robust basis for large-scale exploration of lncRNA-protein interactions and supports future applications in functional genomics and drug discovery.
Related
- Relation-aware multimodal knowledge graph learning for drug–drug interaction prediction
- HAFI-DTA: an interpretable multimodal framework for drug-target affinity prediction based on hierarchical attention guidance and feature iterative interaction
- Contrastive learning and advanced bundle metrics outperform fractional anisotropy for Alzheimer’s disease diffusion MRI prediction: findings from the Beyond FA challenge
- RNA-guided contrastive learning enhances patient-level prediction from histology
- A thermo-responsive collapse system for controlling heterogeneous cell localization, ratio and interaction for three-dimensional solid tumor modeling
BibTeX
@article{zhao2026printer,
title = {{PRInter: an inductive multimodal heterogeneous network embedding framework with contrastive learning for lncRNA-protein interaction prediction}},
author = {Guoqing Zhao and Baolu Shi and Pengpai Li and Zhi-Ping Liu},
journal = {BMC Bioinformatics},
year = {2026},
month = jul,
doi = {10.1186/s12859-026-06571-4},
url = {https://doi.org/10.1186/s12859-026-06571-4},
note = {CC BY-NC-ND 4.0},
}
RIS
TY - JOUR
TI - PRInter: an inductive multimodal heterogeneous network embedding framework with contrastive learning for lncRNA-protein interaction prediction
AU - Guoqing Zhao
AU - Baolu Shi
AU - Pengpai Li
AU - Zhi-Ping Liu
PY - 2026
DA - 2026/07/30
JO - BMC Bioinformatics
DO - 10.1186/s12859-026-06571-4
UR - https://doi.org/10.1186/s12859-026-06571-4
AB - Background: Long non-coding RNAs (lncRNAs) regulate diverse biological processes via protein interactions, and disruptions in these interactions are frequently associated with complex diseases. Network-based computational methods show promise in predicting such interactions, yet most cannot generalize to new, unseen molecules. Results: Here, we introduce PRInter, an inductive multimodal framework integrating sequence features with heterogeneous network structures for lncRNA-protein interaction prediction. PRInter employs meta-path-guided random walks to capture network context and utilizes contrastive learning to refine molecular sequence embeddings, facilitating the prediction of candidate lncRNA-protein pairs involving molecules beyond the training network. To improve the reliability of evaluation, we adopt a degree-balanced negative sampling strategy to reduce potential bias from scale-free network topology. Comprehensive experiments on three public datasets show that PRInter achieves an AUC of 0.882 and performs favorably compared with representative baseline methods. Conclusions: As an inductive and less biased model, PRInter offers a robust basis for large-scale exploration of lncRNA-protein interactions and supports future applications in functional genomics and drug discovery.
SN - 1471-2105
ER -