Trans-m5C: A transformer-based model for predicting 5-methylcytosine (m5C) sites

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• 5-Methylcytosine (m5C) is a crucial RNA modification impacting RNA localization, stability, and translation.
• Current high-throughput sequencing methods for m5C identification are costly, labor-intensive, and time-consuming.
• There is a need for efficient computational approaches to identify m5C sites.

Purpose of the Study:

• To develop a novel deep-learning method for accurate and efficient m5C site prediction.
• To address the limitations of existing computational methods that rely on complex, unavailable hand-crafted features.
• To provide a competitive alternative to current resource-intensive sequencing technologies.

Main Methods:

• Categorization of m5C sites into NSUN2-dependent and NSUN6-dependent types for distinct feature extraction.
• Utilization of transformer neural networks to extract global sequence features.
• Employing a multi-layer perceptron discriminator for m5C site prediction.

Main Results:

• Trans-m5C demonstrates competitive performance compared to baseline and existing methodologies.
• The method was rigorously evaluated on experimentally validated m5C data from human and mouse species.
• The deep-learning approach effectively captures essential features for m5C site prediction.

Conclusions:

• Trans-m5C offers an efficient and accurate computational approach for m5C site identification.
• The method overcomes the drawbacks of traditional sequencing and feature-engineered computational tools.
• This advancement facilitates broader research into the functional roles of m5C modifications in RNA metabolism.