Interpretable Multi-Scale Deep Learning for RNA Methylation Analysis across Multiple Species

Area of Science:

• Computational Biology
• Genomics
• Molecular Biology

Background:

• RNA modifications are vital for cellular regulation but traditional detection methods are inefficient.
• Existing techniques often lack cross-species applicability, hindering comprehensive analysis.
• A versatile computational method is needed for interpretable, multi-species RNA modification studies.

Purpose of the Study:

• To develop a novel computational model for predicting diverse RNA modifications.
• To enable interpretable, sequential-level analysis of RNA modifications across different species.
• To uncover underlying biological grammars and mechanisms of RNA modifications.

Main Methods:

• A multi-scale, biological language-based deep learning model was designed.
• The model was trained and validated on diverse RNA modification datasets.
• Benchmark comparisons and attention weight visualization were employed for analysis.

Main Results:

• The proposed model significantly outperforms existing state-of-the-art methods in predicting various RNA methylation types.
• Cross-species validation confirms the model's robustness and generalizability.
• Attention weight analysis reveals the model's ability to capture functional genomic semantics.

Conclusions:

• The developed deep learning model offers a superior, interpretable approach for predicting RNA modifications across species.
• The findings suggest the existence of
• biological grammars
• which can map methylation patterns and elucidate RNA modification mechanisms.