PhosBoost: Improved phosphorylation prediction recall using gradient boosting and protein language models.

Area of Science:

• Biochemistry
• Computational Biology
• Plant Science

Background:

• Protein phosphorylation is a crucial post-translational modification regulating plant biological processes.
• Current experimental data for plant phosphorylation is limited, hindering comprehensive analysis.
• Accurate prediction of phosphorylation sites is essential for understanding cellular signaling and function.

Purpose of the Study:

• To develop a scalable and accurate machine-learning method for predicting protein phosphorylation sites in plants.
• To compare the performance of the new method against existing prediction tools.
• To assess the cross-species transferability and scalability of the developed model.

Main Methods:

• Developed PhosBoost, a machine-learning approach combining protein language models and gradient-boosting trees.
• Trained PhosBoost on data from the qPTMplants database.
• Compared PhosBoost with PhosphoLingo and DeepPhos, incorporating a sequence-based pairwise alignment step.

Main Results:

• PhosBoost demonstrated superior recall for serine and threonine phosphorylation prediction compared to existing methods.
• PhosBoost successfully predicted tyrosine phosphorylation sites, where other methods failed.
• The inclusion of pairwise alignment improved prediction accuracy for all tested classifiers.
• PhosBoost models showed cross-species transferability and scalability for genome-wide predictions.

Conclusions:

• PhosBoost offers improved recall for protein phosphorylation site prediction in plants, particularly for tyrosine sites.
• The method is scalable for large-scale genome-wide predictions and transferable across plant species.
• PhosBoost provides a valuable tool for advancing plant phosphoproteomics research.