RNA-protein interaction prediction using network-guided deep learning
Problem
Predicting which RNAs bind which proteins is computationally hard, especially for RNAs and proteins with no known data.
Why it matters
RNA-protein interactions underpin gene regulation and disease understanding, so reliable genome-wide prediction matters for biology and medicine.
What the AI did
ZHMolGraph combines a graph neural network with unsupervised large language models to predict RNA-protein interactions, validated on two benchmark datasets.
Results
AUROC 79.8% and AUPRC 82.0% on unknown RNA/protein dataset; improvements of 7.1-28.7% AUROC and 4.6-30.0% AUPRC over other methods
Impact
Benchmark gains support it as a reliable option for genome-wide RNA-protein prediction, per the abstract.
Record
| Venue | Communications Biology |
|---|---|
| Published | 2025 |
| Work type | article |
| Authors | Haoquan Liu, Yiren Jian, Chen Zeng, Yunjie Zhao |
| Citations | 32 (OpenAlex, at import time) |
| References | 50 |
| Model or method | ZHMolGraph (graph neural network + unsupervised large language models) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s42003-025-07694-9 |
| Open access | gold |
| Keywords | RNA-protein interaction, graph neural network, large language model, AUROC, benchmark, SARS-CoV-2 |
Importance rating
significant — thousands to millions affected
RNA-protein interactions underpin gene regulation and disease understanding; genome-wide prediction is a genuinely new capability with broad potential for biology and medicine, though no scale is quantified.