Boosting phosphorylation site prediction with sequence feature-based machine learning.

Area of Science:

• Biochemistry
• Computational Biology
• Bioinformatics

Background:

• Protein phosphorylation is a critical post-translational modification regulating cellular processes.
• Accurate identification of phosphorylation sites is essential for understanding protein function and signaling pathways.

Purpose of the Study:

• To develop an efficient and accurate computational method for predicting phosphate binding sites in protein sequences.
• To provide an interpretable model based on feature engineering and decision trees.

Main Methods:

• Utilized a LightGBM-based computational approach incorporating evolutionary, geometric, sequence environment, and amino acid-specific features.
• Engineered features manually and employed a decision tree-based classification strategy.
• Validated the method on the Phospho.ELM (P.ELM) benchmark dataset comprising 2429 protein sequences.

Main Results:

• Achieved a higher F1 score (0.504) and ROC AUC (0.836) compared to existing methods.
• Demonstrated significantly reduced computation time compared to deep learning frameworks.
• Incorporating deep learning output improved performance (F1 score = 0.546; ROC AUC = 0.849), indicating the method's adaptability.

Conclusions:

• The proposed LightGBM-based method provides an accurate, interpretable, and computationally efficient solution for predicting protein phosphorylation sites.
• The model's interpretability offers insights into the biophysical features governing phosphorylation site recognition.
• The approach serves as a valuable tool for researchers studying protein function and cellular regulation.