PRIDA: Cross-domain protein-RNA interaction prediction based on biological language models and conditional domain-adversarial learning

Motivation The interaction between RNA and proteins is a complex process that involves atom-level details and dynamic changes. Accurate prediction of protein-RNA interactions helps researchers to develop protein-targeted small nucleic acid drugs. However, the scarcity of protein-RNA complexes and experimental data limits the ability of existing data-driven supervised learning methods to learn latent patterns, leading to limited performance. Results In this study, we propose PRIDA (Protein-RNA Interaction prediction based on Domain-Adversarial learning), a protein-RNA interaction framework that combines bilinear attention with conditional domain-adversarial learning. Powered by pretrained protein and RNA language models, PRIDA improves interaction prediction under protein-level distribution shifts and limited labeled data. By integrating PRIDA with Monte Carlo Tree Search, we apply it to a virtual screening pipeline to generate RNA aptamer candidates for specific target proteins. Experimental results show that PRIDA achieves state-of-the-art performance in cross-domain protein-RNA interaction prediction and provides effective guidance for RNA aptamer virtual screening. Availability The data and source code of the proposed method are available at GitHub (https://github.com/zhouyyang/PRIDA) and Zenodo (https://zenodo.org/records/23644677) Supplementary information Supplementary data are available at Bioinformatics online.