IdentPMP

IdentPMP predicts moonlighting proteins in plants to identify plant proteins that perform two or more distinct biological functions.

Key Features:

• Plant-specific dataset: Built from a bespoke dataset of plant protein sequences.
• Redundant-sequence reduction: Reduced redundant sequences to ensure data quality.
• Feature selection: Applied rigorous feature selection to the training data.
• Normalization: Performed normalization on training features.
• Dimensionality reduction: Applied dimensionality reduction to the training data.
• Machine learning evaluation: Evaluated various machine learning methods in preliminary modeling to identify effective feature classes.
• Hyperparameter optimization: Conducted grid search for parameter optimization.
• Core algorithm: Selected eXtreme Gradient Boosting (XGBoost) as the primary prediction algorithm based on performance.
• Independent test performance: Achieved AUPRC of 0.43 and AUC of 0.68 on an independent test set, representing 19.44% and 13.33% improvements respectively over non-plant-specific methods.

Scientific Applications:

• Plant moonlighting protein prediction: Predicts plant proteins that carry out multiple distinct biological functions.
• Functional research support: Supports studies exploring the multifunctional roles of plant proteins.
• Benchmarking and method comparison: Enables comparative evaluation of plant-specific versus non-plant-specific prediction methods.

Methodology:

Constructed a bespoke plant protein dataset with redundant-sequence reduction, applied feature selection, normalization, and dimensionality reduction, performed preliminary modeling with multiple machine learning methods followed by comparative analysis and grid search for parameter optimization, selected XGBoost as the final algorithm, and evaluated performance on an independent test set (AUPRC 0.43, AUC 0.68).