PickPocket

PickPocket predicts ligand-binding pockets on proteins to identify family-specific protein–ligand interactions.

Key Features:

• Ligand-Specific Prediction: Focuses on user-defined sets of ligands to predict binding pockets tailored to specific ligand families.
• Neural Network-Based Approach: Employs neural networks that use pocket descriptors and secondary structure information as inputs.
• High Predictive Accuracy: Achieved greater than 90% prediction accuracy for fatty acid-like ligands using a dataset of 1,740 manually curated ligand-binding pockets.
• Non-linear Motif Detection: Identifies binding-site residues that are not contiguous in the protein sequence.

Scientific Applications:

• Protein Function Discovery: Facilitates discovery of novel protein functions by mapping ligand-specific binding pockets.
• Study of Under-represented Ligands: Enables analysis of ligand families with limited structural representation, such as fatty acid-like molecules.
• Prediction on Unseen Structural Data: Supports prediction of binding sites for newly determined or previously unseen protein structures.

Methodology:

Train neural networks on datasets of known ligand-binding pockets using pocket descriptors and secondary structure information to identify binding sites that may comprise non-contiguous residues.