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

Area of Science:

• Molecular Biology
• Bioinformatics
• Genomics

Background:

• Long non-coding RNAs (lncRNAs) possess dual protein-coding and regulatory functions, termed bifunctional RNAs.
• Accurate identification of bifunctional lncRNAs is crucial for advancing RNA biology and biomarker discovery.
• Current methods for identifying bifunctional lncRNAs require improvement in accuracy and scope.

Purpose of the Study:

• To develop and validate a novel computational model, cncFinder, for accurate prediction of bifunctional lncRNAs.
• To enhance the understanding of RNA multifunctionality and its implications in biological processes.
• To provide a user-friendly tool for researchers to identify bifunctional lncRNAs.

Main Methods:

• Developed cncFinder, a graph-attention-network-based model utilizing k-mer graphs and Word2Vec for feature encoding.
• Employed graph attention networks to capture complex sequence dependencies in lncRNA transcripts.
• Validated cncFinder's performance on independent test datasets and cross-species data (mouse, fruit fly).

Main Results:

• cncFinder demonstrated superior predictive performance compared to state-of-the-art models on testing datasets.
• The model showed robustness and broad applicability across different species.
• Interpretability analysis identified biologically relevant motifs, including start codons and Kozak-like elements, validating its biological relevance.

Conclusions:

• cncFinder significantly advances the accuracy and interpretability of bifunctional lncRNA prediction.
• The tool provides a powerful resource for systematic discovery of bifunctional lncRNAs, offering new insights into RNA multifunctionality.
• A user-friendly web server enhances accessibility for the research community.