AIRBP

AIRBP predicts RNA-binding proteins from sequence data using a stacking-based machine learning approach to enable annotation and study of RNA-related processes.

Key Features:

• Machine learning architecture: Uses a stacking ensemble to combine classifiers for RBP prediction.
• Input features: Derives features from evolutionary information, physicochemical properties, and disordered regions of proteins.
• Feature selection: Employs an evolutionary algorithm (EA) to identify an optimal feature subset for training.
• Evaluation strategy: Validated using 10-fold cross-validation on a benchmark dataset.
• Benchmark performance: Achieves ACC = 95.38%, F1-score = 0.917, and MCC = 0.885 under 10-fold cross-validation on the benchmark dataset.
• Independent test performance: Human test set ACC = 93.04%, F1-score = 0.943, MCC = 0.855; S. cerevisiae test set ACC = 91.60%, F1-score = 0.942, MCC = 0.789; A. thaliana test set ACC = 91.67%, F1-score = 0.953, MCC = 0.594.
• Comparative performance: Reported to outperform existing state-of-the-art methods in RBP prediction.

Scientific Applications:

• RBP annotation: Identification and annotation of RNA-binding proteins from protein sequence data.
• RNA metabolism studies: Supporting analysis of processes including splicing, mRNA stabilization, localization, translation, synthesis, folding, modification, processing, and degradation by identifying associated RBPs.
• Disease and therapeutic research: Facilitating investigation of RBPs implicated in disease biology and potential therapeutic development.

Methodology:

Implements stacking-based machine learning using features from evolutionary information, physicochemical properties, and disordered regions; selects feature subsets with an evolutionary algorithm (EA); training and validation performed with 10-fold cross-validation and independent test sets.