iDBP

iDBP identifies DNA-binding proteins from three-dimensional protein structures by detecting evolutionarily conserved surface patches with the PatchFinder algorithm and classifying proteins using a random forests classifier.

Key Features:

• PatchFinder algorithm: Detects clusters of evolutionarily conserved regions on protein surfaces that are indicative of functional DNA-binding sites.
• Feature extraction: Computes electrostatic potential of identified patches, cluster-based amino acid conservation patterns within patches, secondary structure content of patches, and whole-protein dipole moment.
• Random forests classifier: Trained to distinguish DBPs from non-DNA-binding proteins using three-dimensional structural features.
• Training dataset and performance: Classifier trained on 138 known DBPs and 110 non-DNA-binding proteins, achieving sensitivity 0.90 and specificity 0.90.
• Comparative validation: Tested against five different methods on 11 new DBPs and uniquely annotated all 11 correctly.
• Application to unannotated structures: Applied to 757 proteins of known structure but unknown function, predicting 218 likely DNA binders and suggesting potential novel structural motifs for DNA interaction.

Scientific Applications:

• DBP identification from structural databases: Enables detection of DNA-binding proteins within repositories of three-dimensional protein structures.
• Functional annotation of unknown proteins: Prioritizes proteins of unknown function for experimental validation by predicting DNA-binding capability.
• Discovery of novel binding motifs: Reveals candidate novel protein-DNA interaction motifs to inform studies in genomics, molecular biology, and drug design.

Methodology:

Detect conserved surface patches with the PatchFinder algorithm; extract electrostatic potential, cluster-based amino acid conservation, secondary structure content of patches, and whole-protein dipole moment; train and validate a random forests classifier on 138 DBPs and 110 non-DBPs; perform comparative testing against five methods on 11 new DBPs; apply classifier to 757 structures to predict 218 DNA binders.