DeepMPSF: A Deep Learning Network for Predicting General Protein Phosphorylation Sites Based on Multiple Protein...

Area of Science:

• Biochemistry
• Computational Biology
• Bioinformatics

Background:

• Post-translational modifications, particularly phosphorylation, are crucial for cellular processes and diseases.
• Existing computational methods for predicting phosphorylation sites often lack comprehensive contextual information due to reliance on limited sequence features.

Purpose of the Study:

• To develop an advanced computational model, DeepMPSF, for accurate prediction of protein phosphorylation sites.
• To overcome limitations of existing methods by incorporating diverse protein sequence features.

Main Methods:

• DeepMPSF utilizes two subnetworks (S71SFE and BBFE) to extract sequence semantic and protein background biophysical features.
• The model employs ensemble learning to address imbalanced datasets during training and prediction.

Main Results:

• DeepMPSF demonstrates superior prediction performance for S/T and Y residues compared to benchmark methods on human protein datasets.
• The model exhibits excellent cross-species generalization, significantly improving AUC, F1-score, and MCC metrics on *Mus musculus* and *Rattus norvegicus* test sets.

Conclusions:

• DeepMPSF's multi-feature approach significantly enhances phosphorylation site prediction accuracy.
• The model provides valuable insights for future research in phosphorylation site analysis and downstream applications.