CoCoNat

CoCoNat predicts coiled-coil domains (CCDs) in protein sequences and annotates residue-level registers and oligomerization states for structural and functional characterization.

Key Features:

• Predictive capabilities: Detects coiled-coil helix boundaries, annotates residue-level registers including the heptad repeat pattern, and predicts oligomerization states.
• Sequence encoding: Encodes protein sequences by combining two state-of-the-art protein language models.
• Deep-learning architecture: Implements a three-step deep-learning procedure concatenated with a Grammatical-Restrained Hidden Conditional Random Field (GR-HCRF) for CCD identification and refinement.
• Oligomerization module: Employs a final neural network component dedicated to predicting oligomerization states.
• Performance: Outperforms existing methods on residue-level and segment-level CCD prediction, register annotation, and oligomerization-state prediction on a blind test set.

Scientific Applications:

• Protein functional annotation: Provides residue- and segment-level CCD annotations to support protein functional annotation across organisms.
• Structural characterization: Enables mapping of heptad registers and inference of oligomerization states for coiled-coil structural characterization.

Methodology:

CoCoNat combines sequence encodings from two protein language models, applies a three-step deep-learning procedure concatenated with a Grammatical-Restrained Hidden Conditional Random Field (GR-HCRF) for identification and refinement of CCDs, and uses a final neural network to predict oligomerization states.