DNCON2

DNCON2 predicts protein residue-residue contacts and inter-residue distance distributions using convolutional neural networks to inform protein tertiary structure modeling.

Key Features:

• Deep Learning-Based Prediction: DNCON2 employs convolutional neural networks (CNNs) with a multi-distance approach to capture global coevolutionary coupling patterns across correlated contacts.
• Integration of Multiple Factors: The method integrates multiple sequence alignments (MSA), distance distribution prediction, deep learning, and domain-based contact integration to improve contact accuracy.
• Domain-Based Contact Prediction: DNCON2 incorporates an ab initio approach to parse domains from MSAs and integrate domain-specific contacts without relying on known protein structures.
• Multi-Distance Interval Prediction: The tool predicts inter-residue distances across multiple intervals rather than binary contacts to capture richer structural information.
• Benchmarking and Performance: In CASP13 evaluation on 75 targets comprising 108 domains, DNCON2's CNN-based methods outperformed three coevolution-based methods, increasing precision by 19.2 percentage points.

Scientific Applications:

• Protein Structure Prediction: Provides contact maps and inter-residue distance distributions to guide tertiary structure modeling.
• Protein Folding Studies: Supplies residue-contact information for analysis of folding mechanisms and structural constraints.
• Function and Interaction Inference: Enables inference of functional sites and protein–protein interaction interfaces from spatial residue arrangements.

Methodology:

DNCON2 uses CNN-based multi-distance interval prediction to capture global coevolutionary coupling from MSAs, applies an ab initio domain-parsing approach on MSAs to integrate domain-specific contacts, leverages distance distribution prediction and deeper sequence alignments, and was evaluated in CASP13 on 75 targets (108 domains) showing a 19.2 percentage point precision improvement over three coevolution-based methods.