SAPPHIRE

SAPPHIRE predicts thermophilic proteins from amino acid sequences using a stacking-based ensemble learning framework to improve identification of thermophilic proteins for protein biochemistry and enzyme development.

Key Features:

• Stacking-based ensemble: Implements a stacking ensemble that integrates outputs from multiple baseline models to produce a meta-predictor (SAPPHIRE).
• Feature encodings: Uses twelve distinct feature encodings to represent protein sequences without requiring structural information.
• Baseline models: Trains 72 baseline models by combining twelve feature encodings with six popular machine learning algorithms.
• Genetic algorithm selection: Employs a genetic algorithm together with a self-assessment-report approach to mine and select informative predicted probabilities from baseline models.
• Meta-predictor optimization: Builds and refines the final meta-predictor using an optimal feature set derived from selected predictions.
• Performance evaluation: Achieves 10-fold cross-validation accuracy of 0.942 and Matthew's correlation coefficient (MCC) of 0.884, with reported improvements over existing methods.

Scientific Applications:

• Large-scale TPP identification: Enables high-throughput prediction of thermophilic proteins from sequence data for proteome-scale screens.
• Enzyme development: Supports selection of candidate thermophilic enzymes for biotechnology and industrial applications.
• Protein thermostability research: Facilitates studies in protein biochemistry focused on thermostability determinants using sequence-based predictions.

Methodology:

Train 72 baseline models by combining twelve feature encodings with six machine learning algorithms; apply a genetic algorithm and a self-assessment-report approach to mine and select informative predicted probabilities; construct and optimize a stacking meta-predictor (SAPPHIRE) using an optimal feature set; evaluate performance with 10-fold cross-validation.