Phosphorylation site prediction with a modified k-nearest neighbor algorithm and BLOSUM62 matrix.

Area of Science:

• Bioinformatics
• Computational Biology
• Molecular Biology

Background:

• Protein phosphorylation is a key post-translational modification in eukaryotes.
• Experimental identification of phosphorylation sites and their kinases is laborious and resource-intensive.
• Computational prediction methods offer rapid annotation for guiding experimental validation.

Purpose of the Study:

• To develop and evaluate a novel machine learning approach for predicting protein phosphorylation sites and their specific kinases.
• To improve the accuracy and efficiency of phosphorylation site prediction compared to existing methods.

Main Methods:

• A modified k-Nearest Neighbor (k-NN) algorithm utilizing Manhattan distance was developed.
• BLOSUM62-based similarity scores between phosphorylation sites served as input features.
• The method was validated using leave-one-out testing on protein kinase (PK) groups PKA and CK2.

Main Results:

• The proposed k-NN method demonstrated superior performance in predicting phosphorylation sites.
• The approach outperformed established methods such as Scansite and NetPhosK in leave-one-out cross-validation.
• The findings highlight the effectiveness of sequence-based similarity scores for kinase substrate prediction.

Conclusions:

• The developed k-NN method provides a competitive and efficient computational tool for phosphorylation site prediction.
• This approach can serve as a valuable guideline for experimental studies in proteomics and kinase research.
• The study contributes to advancing bioinformatics tools for understanding protein post-translational modifications.