DiANNA

DiANNA predicts the oxidation states and disulfide connectivity of cysteines within protein sequences and classifies cysteine binding to metallic ligands (iron, zinc, cadmium, carbon) to inform protein structure and function analysis.

Key Features:

• Cysteine Oxidation State Prediction: A feed-forward neural network predicts whether cysteines are in a reduced sulfhydryl state or involved in disulfide bonds (half-cystines).
• Disulfide Connectivity Prediction: A diresidue neural network predicts which cysteine pairs form disulfide bonds by analyzing symmetric flanking regions and incorporating secondary structure and evolutionary information.
• Metallic Ligand Binding Prediction (v1.1): A support vector machine with a spectrum kernel predicts cysteine binding to metallic ligands, distinguishing iron, zinc, cadmium, and carbon.
• Integration of Secondary Structure and Evolutionary Information: Secondary structure predictions from PSIPRED and evolutionary profiles from multiple sequence alignments using PSIBLAST against the non-redundant SwissProt database are used as inputs.
• Graph-theoretic Matching of Predicted Pair Scores: Rothberg's implementation of Gabow's maximum weighted matching algorithm is applied to diresidue neural network scores to produce final connectivity predictions.

Scientific Applications:

• Disulfide Bond Topology Determination: Prediction of disulfide connectivity supports tertiary structure modeling and constraint-based structure prediction.
• Protein Structural and Functional Analysis: Cysteine state classification aids interpretation of protein folding, stability, and function related to disulfide bonds and metal coordination.
• Applications in Structural Biology, Biochemistry, and Molecular Medicine: Predictions inform studies of protein structure–function relationships and investigations of cysteine-mediated biochemical mechanisms.

Methodology:

Secondary structure prediction with PSIPRED and evolutionary information from PSIBLAST searches against the non-redundant SwissProt database are used to train a feed-forward neural network for cysteine oxidation states; a diresidue neural network processes symmetric flanking regions augmented with secondary structure and evolutionary data to score cysteine pairings; an SVM with a spectrum kernel performs metallic ligand binding prediction (v1.1); Rothberg's implementation of Gabow's maximum weighted matching algorithm is applied to diresidue neural network scores to derive final disulfide connectivity.