A CNN based m5c RNA methylation predictor.

Area of Science:

• Molecular Biology
• Bioinformatics
• Genomics

Background:

• Post-transcriptional RNA modifications are crucial for biological processes.
• N6-methyladenosine (m6A) and 5-methylcytosine (m5c) are key RNA modifications impacting gene expression.
• Experimental detection of m5c sites is labor-intensive and costly.

Purpose of the Study:

• To develop an efficient computational model for identifying m5c methylation sites in RNA.
• To analyze RNA sequences of full length, not limited to a central motif.
• To overcome limitations of conventional methods in handling high-dimensional RNA sequence data.

Main Methods:

• An end-to-end, 1D Convolutional Neural Network (CNN) model was employed.
• The model was trained and evaluated on pre-processed RNA sequences (41 nucleotides) and full-length sequences.
• Feature extraction techniques were optimized for direct RNA sequence processing.

Main Results:

• The proposed 1D CNN model achieved high performance in classifying m5c methylated sites.
• Achieved 96.70% sensitivity and 96.21% accuracy for 41-nucleotide sequences.
• Demonstrated 96.10% accuracy for full-length RNA sequences, outperforming existing methods.

Conclusions:

• The 1D CNN model offers a robust and accurate approach for m5c site identification.
• The model's ability to process full-length sequences provides a more comprehensive analysis.
• This computational method significantly advances RNA modification research by reducing experimental burden.