TargetATPsite

TargetATPsite predicts Adenosine-5'-triphosphate (ATP) binding sites in protein sequences to identify protein-ligand interactions for protein function annotation and drug discovery.

Key Features:

• Template-Free Prediction: A sequence-based template-free approach predicts ATP binding sites without relying on pre-existing structural templates.
• Binding Residues Prediction: Uses an image sparse representation technique to encode residue evolution information as input features for residue-level binding prediction.
• Binding Pockets Identification: Applies a spatial clustering algorithm to predicted binding residues to identify binding pockets and localize binding regions.
• Machine Learning Approach: An ensemble classifier based on support vector machines (SVM) with multiple random under-samplings addresses imbalance between positive and negative training samples.
• Benchmarking Validation: Experimental validation on three benchmark datasets demonstrates performance that outperforms existing ATP-specific sequence-based predictors by integrating residue prediction and pocket identification.

Scientific Applications:

• Protein Function Annotation: Identification of ATP binding sites supports annotation of protein functions by revealing potential ligand-binding residues.
• Drug Discovery: Pinpointing ATP binding pockets enables targeting of protein-ATP interactions relevant to therapeutic development.

Methodology:

Sequence-based template-free prediction using image sparse representation of residue evolutionary information as features, an ensemble SVM classifier with multiple random under-samplings for residue prediction, and spatial clustering of predicted residues to identify binding pockets.