EMDLP: Ensemble multiscale deep learning model for RNA methylation site prediction.

Area of Science:

• Epitranscriptomics and RNA modifications
• Computational biology and bioinformatics
• Genomics and molecular biology

Background:

• RNA modifications are crucial for gene regulation, but experimental identification is challenging.
• Machine learning offers efficient computational approaches for RNA sequence analysis.
• Existing deep learning models like CNNs and LSTMs have limitations in capturing both local and sequential RNA features.

Purpose of the Study:

• To develop a novel deep learning framework for accurate identification of RNA methylation sites.
• To integrate natural language processing (NLP) techniques with deep learning (DL) for enhanced RNA sequence representation.
• To overcome the limitations of individual CNN and LSTM models in RNA modification site prediction.

Main Methods:

• An ensemble multiscale deep learning predictor (EMDLP) was developed.
• RNA sequences were encoded using NLP methods: RNA word embedding, One-hot encoding, and RGloVe.
• A dilated convolutional Bidirectional LSTM network (DCB) model was employed for feature extraction.
• A soft voting strategy integrated the results from different encoding methods.

Main Results:

• The EMDLP achieved high predictive performance for m1A and m6A methylation sites.
• AUROC values of 95.56% for m1A and 85.24% for m6A were obtained, outperforming state-of-the-art models.
• A user-friendly webserver was made publicly available for EMDLP.

Conclusions:

• A novel predictor, EMDLP, was successfully developed for identifying m1A and m6A methylation sites.
• The study highlights the effectiveness of combining NLP and DL for RNA modification prediction.
• The developed tool provides a valuable resource for epitranscriptomic research.