Cross-RNA transferable sequence representation learning for lncRNA m6A site detection via novel deep domain...

Area of Science:

• Epitranscriptomics
• Computational Biology
• Genomics

Background:

• N6-methyladenosine (m6A) is a critical epitranscriptomic modification found in long noncoding RNAs (lncRNAs), influencing complex disease mechanisms.
• Accurate identification of m6A sites in lncRNAs is vital for disease research, but computational prediction is hindered by limited annotated data and poor generalizability of existing messenger RNA (mRNA)-focused tools.
• Leveraging shared features between lncRNAs and mRNAs is essential for developing effective lncRNA m6A site predictors.

Purpose of the Study:

• To develop a robust computational framework, DSNm6A, for accurate prediction of m6A sites in lncRNAs.
• To create a transferable learning approach that utilizes sequence representations common to both lncRNAs and mRNAs.
• To enhance the generalizability and accuracy of m6A site detection across different RNA types and species.

Main Methods:

• DSNm6A employs a deep learning architecture integrating CNN, Bi-LSTM, and BERT modules for sequence encoding.
• RNA sequences (lncRNA and mRNA) are encoded using multiple facets: One-Hot encoding, nucleotide physicochemical properties, cumulative frequency, and position-specific propensity.
• A domain separation network disentangles domain-invariant (shared) features from domain-specific features to improve prediction accuracy.

Main Results:

• DSNm6A significantly outperforms existing methods in predicting lncRNA m6A sites across various performance metrics.
• The framework demonstrates superior capacity in learning transferable m6A-related sequence features applicable across different RNA types.
• DSNm6A exhibits strong robustness and generalization capabilities across different species, indicating broad applicability.

Conclusions:

• DSNm6A provides an effective solution for lncRNA m6A site prediction by learning cross-RNA transferable features.
• The proposed domain separation network successfully identifies shared sequence patterns crucial for accurate m6A site detection in lncRNAs.
• DSNm6A represents a significant advancement in epitranscriptomic research, facilitating a deeper understanding of lncRNA functions in complex diseases.