PRMxAI: protein arginine methylation sites prediction based on amino acid spatial distribution using explainable...

Area of Science:

• Biochemistry
• Computational Biology
• Bioinformatics

Background:

• Protein methylation is a key post-translational modification regulating cellular functions.
• Arginine methylation is vital for processes like gene regulation and signal transduction.
• Experimental prediction methods are costly and time-consuming.

Purpose of the Study:

• To develop a novel computational method for predicting arginine methylation sites.
• To establish an efficient and accurate alternative to experimental prediction techniques.

Main Methods:

• Developed PRMxAI, a machine learning-based predictor.
• Extracted sequence-based features: dipeptide composition, physicochemical properties, amino acid composition, and information theory-based features.
• Utilized Random Forest as the core classification algorithm.
• Employed 10-fold cross-validation for performance evaluation.

Main Results:

• PRMxAI achieved 87.17% accuracy for mono-methylarginine and 90.40% for di-methylarginine prediction.
• Feature importance was analyzed using explainable artificial intelligence (AI).
• The Random Forest classifier demonstrated superior performance.

Conclusions:

• PRMxAI effectively predicts arginine methylation sites.
• Explainable AI confirmed the model's predictive mechanism.
• PRMxAI outperforms existing state-of-the-art prediction tools.