Probabilistic Prediction of Protein Phosphorylation Sites Using Classification Relevance Units Machines.

Area of Science:

• Biochemistry
• Computational Biology
• Bioinformatics

Background:

• Protein phosphorylation is a crucial post-translational modification regulating cellular processes.
• Experimental identification of phosphorylation sites is costly and time-consuming.
• Computational prediction of phosphorylation sites is highly desirable.

Purpose of the Study:

• To introduce a novel kernel-based probabilistic method, Classification Relevance Units Machine (CRUM), for *in silico* phosphorylation site prediction.
• To evaluate CRUM's performance and efficiency compared to existing methods like Support Vector Machine (SVM).
• To explore the impact of different features (BLOSUM encoding, disorder, amino acid composition) on CRUM's predictive accuracy.

Main Methods:

• Development of a CRUM model for predicting protein phosphorylation sites.
• Implementation of CRUM using various feature combinations: BLOSUM encoding, protein disorder, and amino acid composition.
• Evaluation of CRUM predictors using cross-validation and benchmark experiments.

Main Results:

• CRUM demonstrates predictive performance comparable to SVM but offers a more parsimonious model.
• CRUM exhibits lower training run-time and memory complexity than Relevance Vector Machine (RVM).
• CRUM predictors incorporating BLOSUM encoding achieved top performance in cross-validation and benchmark tests.

Conclusions:

• CRUM is a viable and efficient computational tool for predicting protein phosphorylation sites.
• The BLOSUM feature significantly enhances CRUM's predictive capabilities.
• Further improvements in CRUM predictors are possible through comparative studies with existing tools.