Pf-Phospho: a machine learning-based phosphorylation sites prediction tool for Plasmodium proteins.

Area of Science:

• Bioinformatics
• Computational Biology
• Parasitology

Background:

• Existing in silico tools for phosphosite prediction lack coverage for Plasmodium proteins.
• Advances in machine learning and available phosphoproteomics data present an opportunity to develop predictive models for Plasmodium.

Purpose of the Study:

• To develop and validate an ML-based method for predicting protein phosphorylation sites in Plasmodium.
• To create a bioinformatics resource for analyzing Plasmodium phosphosignaling networks.

Main Methods:

• Development of Pf-Phospho using Random Forest classifiers trained on a dataset of 12,096 Plasmodium falciparum and Plasmodium bergei phosphosites.
• Training/validation using 75% of the data, with the remaining 25% reserved for blind testing.
• Integration with existing resources like PlasmoDB, MPMP, Pfam, and AlphaFold2 predicted structures.

Main Results:

• Pf-Phospho achieves 84% sensitivity, 75% specificity, and 78% precision for kinase-independent phosphosite prediction.
• The tool accurately predicts kinase-specific phosphosites for five key Plasmodium kinases.
• Pf-Phospho demonstrates superior performance compared to tools trained on mammalian phosphoproteome data.

Conclusions:

• Pf-Phospho is the first bioinformatics resource for ML-based prediction of Plasmodium phosphosignaling networks.
• The tool provides a user-friendly platform for integrated analysis of signaling, metabolic, and protein-protein interaction networks in Plasmodium.
• This resource is valuable for advancing research into the phosphorylation patterns of malaria parasites.