A Review of Machine Learning and Algorithmic Methods for Protein Phosphorylation Site Prediction

Area of Science:

• Biochemistry and Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• Post-translational modifications (PTMs) diversify protein functions and regulate cellular processes.
• Phosphorylation is a critical PTM involved in numerous biological pathways, with dysregulation linked to diseases like cancer and neurological disorders.

Purpose of the Study:

• To consolidate and review existing knowledge on phosphorylation site (p-site) prediction methods.
• To evaluate the performance of current p-site prediction tools on novel protein data.

Main Methods:

• Comprehensive review of databases, dataset creation, preprocessing, and evaluation metrics for p-site prediction.
• Investigation of algorithmic and machine learning (ML) based computational methods, including conventional and deep learning approaches.
• Analysis of feature extraction techniques used in p-site prediction.

Main Results:

• Current online p-site prediction tools exhibit notably lower performance on unseen proteins (dbPTM 2022 release) compared to their reported results.
• The performance gap is evident when tested on proteins not included in earlier releases (e.g., dbPTM 2019).

Conclusions:

• Existing computational tools for p-site prediction have limitations in generalizing to new protein data.
• Further research is needed to improve the accuracy and reliability of p-site prediction methods for biological applications.