CFPLM: Improve Protein-RNA Interaction Prediction with a Collaborative Framework Powered by Language Models

Protein-RNA interactions (PRIs) play pivotal roles in biological processes such as gene regulation, making their prediction essential for therapeutic and mechanistic studies. While traditional wet-lab methods are timeconsuming and challenging, computational approaches offer efficient alternatives. Graph-based methods show promise by capturing both direct interactive domains (protein-RNA interaction) and indirect collaborative domains (functional similarity among proteins/RNAs). However, integrating these domains and learning meaningful node representations remain critical challenges. To address this, we propose CFPLM, a collaborative framework fusing large language models, graph convolutional networks, and cross-attention mechanisms to improve PRI prediction. Experiment results demonstrate that CFPLM achieves robust, state-of-the-art performance across three benchmark datasets. It's anticipated to have applicability to similar other interaction prediction tasks. The data and codes are available at: https://github.com/HuanchaoFeng/CFPLM.