BERMP: a cross-species classifier for predicting m⁶A sites by integrating a deep learning algorithm and a random...

Area of Science:

• Genomics and Bioinformatics
• Molecular Biology
• Computational Biology

Background:

• N-methyladenosine (m6A) is a crucial RNA modification impacting various biological processes.
• Existing m6A prediction tools are often species-specific and utilize conventional machine learning.
• High-throughput experiments have generated extensive m6A site data, enabling computational prediction development.

Purpose of the Study:

• To develop a novel, cross-species deep-learning classifier for predicting m6A sites.
• To address the limitations of existing species-centric m6A predictors.
• To create a robust and accurate tool for identifying m6A sites computationally.

Main Methods:

• Development of a deep-learning classifier using Bidirectional Gated Recurrent Unit (BGRU).
• Integration of BGRU with a random forest classifier and enhanced nucleic acid content encoding.
• Creation of the BGRU-based Ensemble RNA Methylation site Predictor (BERMP).

Main Results:

• BGRU demonstrated strong performance on large mammalian datasets but was less effective on smaller yeast datasets.
• The integrated BERMP approach compensated for BGRU's data size sensitivity, showing competitive performance.
• BERMP outperformed existing m6A predictors in both cross-validation and independent tests across different species.

Conclusions:

• BERMP is a novel, high-confidence, multi-species tool for m6A site identification.
• The developed classifier offers improved accuracy and broader applicability compared to existing methods.
• BERMP is freely available, facilitating advancements in RNA methylation research.