AttenRNA: multi-scale deep attentive model with RNA feature variability analysis.

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• Accurate RNA identification is crucial for understanding gene regulation and disease.
• Current methods often focus on binary classification, missing complex cross-type sequence patterns.
• Distinguishing messenger RNAs (mRNAs), long non-coding RNAs (lncRNAs), and circular RNAs (circRNAs) is challenging.

Purpose of the Study:

• To develop a multi-class classification model for simultaneous differentiation of diverse RNA types.
• To address limitations of existing binary classification approaches in RNA identification.
• To enable systematic RNA function analysis through improved classification.

Main Methods:

• Developed AttenRNA, a multi-class classification model.
• Integrated multi-scale k-mer embeddings and attention mechanisms.
• Utilized Uniform Manifold Approximation and Projection for dimensionality reduction.

Main Results:

• AttenRNA achieved high weighted F1 scores of 89.8% (validation) and 89.6% (test).
• Demonstrated robust classification performance and ability to learn discriminative RNA features.
• Showed strong generalization on cross-species data (mouse RNA: 83.89% validation, 83.38% test).

Conclusions:

• AttenRNA provides a reliable and scalable solution for multi-class RNA identification.
• The model effectively differentiates various RNA types, including mRNAs, lncRNAs, and circRNAs.
• AttenRNA facilitates deeper insights into RNA functions and roles in biological processes.