DeepPRMS: advanced deep learning model to predict protein arginine methylation sites

Area of Science:

• Biochemistry and Molecular Biology
• Computational Biology
• Bioinformatics

Background:

• Protein methylation is a critical post-translational modification influencing cellular processes like transcription and DNA repair.
• Experimental methods for identifying methylation sites are often time-consuming and expensive.
• Accurate prediction of protein methylation sites is essential for advancing biological research and drug discovery.

Purpose of the Study:

• To develop a novel, accurate in silico method for predicting protein methylation sites.
• To introduce DeepPRMS, a deep learning-based predictor for identifying protein methylation sites from primary sequences.

Main Methods:

• Utilized deep learning, combining Gated Recurrent Unit (GRU) for sequential information and Convolutional Neural Network (CNN) for spatial information.
• Integrated latent representations from GRU and CNN models for enhanced feature interaction.
• Developed a novel predictor named DeepPRMS for protein methylation site identification.

Main Results:

• DeepPRMS achieved high performance on an independent test dataset, with an accuracy of 85.32%.
• The predictor demonstrated strong predictive capabilities with a specificity of 84.94% and sensitivity of 85.80%.
• Achieved a Matthew's correlation coefficient of 0.71, indicating robust performance and outperforming existing state-of-the-art models.

Conclusions:

• DeepPRMS offers a highly accurate and efficient computational approach for predicting protein methylation sites.
• The tool is expected to significantly guide experimental research in identifying potential methylated protein sites.
• The developed web server provides accessible utility for the scientific community to identify methylation sites.