cncFinder: a graph attention network-based interpretable learning model to identify bifunctional long non-coding RNAs

The discovery of RNAs that possess both protein-coding and regulatory functions, known as bifunctional RNAs, has broadened our understanding of RNA biology. Long non-coding RNAs, which play crucial roles in gene regulation and cellular processes, represent an important subset of such molecules. Accurate and systematic identification of bifunctional long non-coding RNAs is essential for uncovering their roles in physiology and disease, and for advancing biomarker discovery and therapeutic development. In this study, we present cncFinder, a deep learning model based on graph attention networks designed to predict bifunctional long non-coding RNAs. The framework converts RNA sequences into k-mer graphs, encodes node features with Word2Vec, and applies graph attention mechanisms to capture higher-order sequence dependencies. On benchmark datasets, cncFinder achieved superior performance compared with existing approaches and demonstrated strong robustness across species, including mouse and fruit fly. Interpretability analysis indicated that the model recognized biologically relevant motifs, such as canonical start codons and Kozak-like elements. In a case study, cncFinder accurately detected a translation initiation motif in LINC00961, consistent with experimental validation. To promote broad accessibility, we developed a user-friendly web server (http:\/\/i-health.info\/cncFinder\/) that allows researchers to perform large-scale predictions. Overall, cncFinder improves predictive accuracy and interpretability, offering a reliable tool for systematic discovery of bifunctional long non-coding RNAs. It provides new perspectives for understanding the molecular mechanisms underlying RNA multifunctionality. The datasets and source code are freely available at https:\/\/github.com\/TangQiang0701\/cncFinder.