pHisPred

pHisPred predicts histidine phosphorylation (pHis) sites from protein sequences to support investigation of protein histidine phosphorylation in signaling pathways and cellular processes.

Key Features:

• Input data: Predicts pHis sites from protein sequences and provides separate models for eukaryotic and prokaryotic proteins.
• Training dataset: Trained on the largest collection of experimentally validated pHis sites assembled for benchmarking.
• Machine learning model: Implements a weighted Support Vector Machine with a Radial Basis Function (SVM-RBF) kernel as the primary classifier.
• Model comparison: Benchmarking compared SVM-RBF performance against Logistic Regression (LR), K-Nearest Neighbors (KNN), Random Forest (RF), and Multi-Layer Perceptron (MLP).
• Validation strategy: Evaluated using randomized tenfold cross-validation on training sets.
• Feature selection: Reduced an initial pool of over 10,000 features to 140 features for the eukaryotic model and 150 features for the prokaryotic model.
• Performance metrics: Reported Area Under the Curve (AUC) and F1-score values of 0.81 and 0.40 for eukaryotic datasets, and 0.78 and 0.46 for prokaryotic datasets during tenfold cross-validation.

Scientific Applications:

• Signal transduction and cellular processes: Facilitates identification of histidine phosphorylation sites to support studies of signaling pathways and cellular functions involving pHis.
• Hypothesis generation and experimental design: Provides predicted pHis sites to inform experimental prioritization and design.

Methodology:

Trained on an extensive experimentally validated pHis dataset using a weighted SVM with RBF kernel, compared to LR, KNN, RF, and MLP, evaluated by randomized tenfold cross-validation, and applying feature selection to reduce >10,000 features to 140 (eukaryotic) and 150 (prokaryotic) features.