CoCoPRED

CoCoPRED predicts structural features of coiled-coil proteins, simultaneously identifying coiled-coil domains (CCD), oligomeric state, and register for structural analysis.

Key Features:

• Deep learning architecture: Uses convolutional layers, bidirectional long short-term memory (BiLSTM) networks, and an attention mechanism.
• Three specialized networks: Comprises distinct but interconnected CCD network, oligomeric state network, and register network.
• Simultaneous multi-feature prediction: Predicts coiled-coil domain (CCD), oligomeric state, and register within a single framework.
• Attention-driven interpretability: Employs attention to identify registers that influence predictions, with registers a, b, and e highlighted as particularly important.
• Cross-validation performance: Demonstrated superior performance for CCD prediction and oligomeric state prediction in 5-fold cross-validation experiments.
• Network interplay: Leverages the relationship between networks such that accurate CCD predictions correlate with reliable oligomeric state predictions.

Scientific Applications:

• CCD identification: Detection of coiled-coil domains in protein sequences.
• Oligomeric state prediction: Determination of coiled-coil oligomeric state from sequence-derived features.
• Register assignment: Assignment of register positions within coiled-coil sequences.
• Register importance analysis: Identification of critical registers (a, b, e) that influence oligomeric state prediction.

Methodology:

CoCoPRED employs a deep neural network architecture incorporating convolutional layers, BiLSTM networks, and an attention mechanism, organized into CCD, oligomeric state, and register networks; models were evaluated using 5-fold cross-validation, and attention weights were analyzed to identify registers a, b, and e as influential while CCD predictions inform oligomeric state prediction.