iDBPs

iDBPs predicts DNA-binding proteins from protein three-dimensional (3D) structures by using evolutionary profiles to identify functional regions and assess DNA-binding capability.

Key Features:

• Input data: Operates on protein three-dimensional (3D) structures and leverages evolutionary profiles.
• Functional region prediction: Identifies functionally significant regions based on conservation patterns and large clusters of conserved residues.
• Feature calculation: Computes average surface electrostatic potential, dipole moment, cluster-based amino acid conservation patterns, predicted functional-region features, and global protein features.
• Classification algorithm: Uses a random forests classifier to determine the likelihood that a protein binds DNA and to provide an estimate of prediction confidence.
• Performance evaluation: Trained and tested on diverse datasets and reported an area under the ROC curve (AUC) of 0.90 on a dataset reflecting the proportion of DBPs in a proteome.
• Database application: Applied to the N-Func database of proteins with solved 3D structures but unknown functions to identify putative DNA-binding proteins.

Scientific Applications:

• Proteome-wide DBP prediction: Identify and quantify DNA-binding proteins within proteomes based on structural and evolutionary features.
• Experimental target prioritization: Generate putative DBPs for experimental validation.
• Functional annotation: Annotate proteins of unknown function in structural databases such as N-Func.
• Protein–DNA interaction studies: Support investigations into protein–DNA interactions by highlighting candidate DNA-binding regions and proteins.

Methodology:

Predict functional regions using evolutionary profiles and conserved-residue clusters; compute features including average surface electrostatic potential, dipole moment, and cluster-based amino acid conservation patterns for predicted functional regions and global protein properties; classify proteins with a random forests model to estimate DNA-binding likelihood and prediction confidence, with training/testing yielding an AUC of 0.90 on a proteome-proportion dataset.