TargetATPsite: a template-free method for ATP-binding sites prediction with residue evolution image sparse...

Area of Science:

• Biochemistry
• Computational Biology
• Bioinformatics

Background:

• Protein-ligand interactions are crucial for understanding protein functions and developing new drugs.
• Identifying Adenosine-5 '-triphosphate (ATP) binding sites is essential for both protein annotation and drug discovery efforts.

Purpose of the Study:

• To develop a novel, sequence-based, template-free predictor called TargetATPsite for identifying ATP binding sites.
• To enhance the prediction of ATP binding sites by incorporating a method for identifying binding pockets.

Main Methods:

• Utilized machine learning approaches, specifically an ensemble classifier of Support Vector Machines (SVMs) with random under-sampling.
• Developed a novel image sparse representation technique to encode residue evolution information for binding residue prediction.
• Implemented a spatial clustering algorithm to identify binding pockets from predicted binding residues.

Main Results:

• The TargetATPsite predictor demonstrated efficacy in identifying ATP binding sites across three benchmark datasets.
• The predictor successfully identified both binding residues and subsequently the binding pockets.
• The proposed methods outperformed existing ATP-specific sequence-based predictors.

Conclusions:

• TargetATPsite offers an effective, sequence-based, template-free approach for predicting ATP binding sites.
• The two-step prediction process, including pocket identification, enhances the accuracy and utility of binding site prediction.
• This tool has significant implications for accelerating drug discovery and protein function annotation.