Deepm6A-MT: A deep learning-based method for identifying RNA N6-methyladenosine sites in multiple tissues.

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• N6-methyladenosine (m6A) is a critical and abundant internal modification in eukaryotic messenger RNA (mRNAs).
• Existing m6A detection methods require improvement in predictive accuracy and efficiency.

Purpose of the Study:

• To propose an enhanced computational method for predicting m6A modification sites.
• To improve the accuracy and efficiency of m6A site prediction.

Main Methods:

• Developed Deepm6A-MT, integrating bi-directional gated recurrent units (Bi-GRU) and convolutional neural networks (CNN).
• Employed dual input channels: one with embedding layer, Bi-GRU, and CNN; the other with various encoding schemes (one-hot, dinucleotide, chemical properties).
• Validated performance using 5-fold cross-validation, independent testing, and cross-species/cross-tissue analyses.

Main Results:

• Deepm6A-MT achieved state-of-the-art performance in predicting m6A modification sites.
• Demonstrated effectiveness and efficiency across different species and tissue types.
• The method's performance surpasses existing m6A prediction tools.

Conclusions:

• Deepm6A-MT offers a highly accurate and efficient approach for m6A site prediction.
• The tool is accessible via a web server (http://www.biolscience.cn/Deepm6A-MT/) to facilitate academic research.
• This advancement aids in understanding the functional roles of m6A modifications in cellular processes.