RBProkCNN

RBProkCNN: Convolutional neural network-based predictor of prokaryotic RNA-binding proteins

RBProkCNN predicts RNA-binding proteins (RBPs) in prokaryotes using machine learning models optimized for prokaryotic sequence characteristics and evolutionary information.

Key Features:

• Integrated Machine Learning Framework: Combines eight shallow learning models and four deep learning frameworks for prokaryotic RBP classification.
• PSSM-Based Evolutionary Features: Utilizes Position-Specific Scoring Matrix (PSSM) profiles to encode contextual evolutionary information.
• CNN Architecture with Feature Prioritization: Implements a convolutional neural network (CNN) enhanced by variable importance measures from extreme gradient boosting to prioritize evolutionarily significant features.
• High Predictive Performance: Achieves 98.04% auROC and 98.19% auPRC in five-fold cross-validation, and 95.77% auROC and 95.78% auPRC in independent validation.

Scientific Applications:

• Prokaryotic RBP Identification: Enables genome-scale prediction of RNA-binding proteins involved in post-transcriptional regulation, mRNA stability, and environmental adaptation in prokaryotes.

Methodology:

Protein sequences are encoded using PSSM-derived evolutionary features. Multiple shallow and deep learning models are trained and integrated, with a CNN classifier refined by extreme gradient boosting-based variable importance analysis to enhance discrimination of prokaryotic RNA-binding proteins.