rBPDL

rBPDL predicts RNA-binding proteins by integrating convolutional neural networks and long short-term memory networks with ensemble voting to perform multilabel classification of RBPs.

Key Features:

• CNN + LSTM architecture: Combines convolutional neural network (CNN) feature extraction with long short-term memory (LSTM) networks to capture spatial and temporal patterns in protein sequence data.
• Multilabel classification: Performs multilabel classification to assign multiple RNA-binding roles or labels to proteins.
• Ensemble voting: Uses an ensemble learning approach with a voting algorithm to aggregate predictions from multiple models for improved robustness.
• Benchmark performance: Achieved macro-AUC 0.936, micro-AUC 0.962, and weighted AUC 0.946 on the RBP68 dataset.
• Domain-level evaluation: Includes validation across RBP domain analyses to assess performance consistency across domain characteristics.
• Amino-acid and physicochemical analysis: Examines amino-acid preferences and physicochemical properties associated with protein–RNA binding.

Scientific Applications:

• RBP identification: Prediction and annotation of RNA-binding proteins from protein sequence data.
• Benchmarking and model comparison: Performance evaluation on benchmark datasets such as RBP68 and RBP86.
• Domain analysis: Assessment of RBP detection consistency across protein domains.
• Binding determinant analysis: Investigation of amino-acid preferences and physicochemical properties that influence protein–RNA interactions.
• Experimental guidance: Informing experimental design and biological research on RBPs through predictive insights.

Methodology:

Integration of convolutional neural network (CNN) with long short-term memory (LSTM) networks for multilabel classification, ensemble learning via a voting algorithm, validation on RBP68 and RBP86, RBP domain analysis, and analysis of amino-acid preferences and physicochemical properties.