An improved residual network using deep fusion for identifying RNA 5-methylcytosine sites

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• 5-Methylcytosine (m5C) is a vital post-transcriptional RNA modification involved in numerous cellular processes.
• Traditional experimental methods for m5C identification are laborious and time-consuming.
• Computational approaches, particularly deep learning, offer efficient alternatives for predicting m5C sites.

Purpose of the Study:

• To develop a robust deep learning model for accurate identification of m5C sites in RNA sequences.
• To leverage a deep fusion approach integrating diverse sequence features for enhanced prediction accuracy.
• To provide a reliable computational tool for m5C site prediction, overcoming limitations of experimental methods.

Main Methods:

• RNA sequences were analyzed using Kmer, K-tuple nucleotide frequency component (KNFC), Pseudo dinucleotide composition (PseDNC), and Physical and chemical property (PCP) for feature extraction.
• Bidirectional long- and short-term memory (BiLSTM) and attention mechanisms were employed to fuse extracted sequence features.
• An improved residual network was utilized for the final classification of m5C sites.
• Model performance was validated using 10-fold cross-validation and independent set testing.

Main Results:

• The deep fusion model achieved high prediction accuracies: 91.87% on Arabidopsis thaliana training sets and 95.55% on M.musculus training sets.
• Independent test sets showed excellent performance with accuracies of 92.27% for Arabidopsis thaliana and 95.60% for M.musculus.
• The model demonstrated a considerable improvement over existing methods, highlighting its robust performance.

Conclusions:

• The developed deep fusion model with an improved residual network is highly effective for predicting m5C sites in RNA.
• The integration of multiple feature extraction methods and deep learning architectures significantly enhances prediction accuracy.
• This computational approach provides a valuable tool for advancing research on m5C modifications and their biological roles.