CNN-Meth: A Tool to Accurately Predict Lysine Methylation Sites Using Evolutionary Information-Based Protein Modeling

Area of Science:

• Biochemistry
• Computational Biology
• Bioinformatics

Background:

• Lysine methylation is a critical posttranslational modification impacting protein function.
• Dysregulation of lysine methyltransferases is associated with diseases like cancer and developmental disorders.
• Precise identification of lysine methylation sites is vital for diagnostics and therapeutics.

Purpose of the Study:

• To introduce CNN-Meth, a novel web-based tool for predicting lysine methylation sites.
• To utilize convolutional neural networks (CNNs) for automated feature extraction in methylation site prediction.
• To improve prediction accuracy compared to traditional methods.

Main Methods:

• Development of CNN-Meth, a web utility employing CNNs for lysine methylation site prediction.
• Leveraging evolutionary, structural, and physicochemical data with binary encoding for model training.
• Utilizing protein modeling for feature extraction, analogous to Protein Language Models (PLMs).
• Automated feature extraction via CNNs to minimize information loss.

Main Results:

• CNN-Meth achieved high prediction accuracy: 96.0% Accuracy, 85.1% Sensitivity, 96.4% Specificity.
• A Matthew's Correlation Coefficient (MCC) of 0.65 was obtained, indicating robust performance.
• The study highlights the potential of PLMs in predicting methylation sites.

Conclusions:

• CNN-Meth offers a highly accurate and automated approach for identifying lysine methylation sites.
• The tool provides a valuable resource for researchers and clinicians in disease diagnosis and treatment.
• The findings support the efficacy of CNN-based methods and suggest PLMs as a future research direction.