Sequence-based predictor of ATP-binding residues using random forest and mRMR-IFS feature selection.

Area of Science:

• Computational Biology
• Biochemistry
• Bioinformatics

Background:

• Predicting adenosine triphosphate (ATP)-binding residues is crucial for understanding protein function and designing drugs.
• Existing methods for predicting ATP-binding sites often face limitations in accuracy and feature representation.

Purpose of the Study:

• To develop a novel computational and statistical approach for accurately predicting ATP-binding residues from protein amino acid sequences.
• To introduce and evaluate a new hybrid feature incorporating predicted secondary structure and orthogonal binary vectors.

Main Methods:

• A random forest model was employed for prediction.
• A novel hybrid feature, PSSMPP (predicted secondary structure and orthogonal binary vectors), was developed.
• The mRMR-IFS feature selection method was utilized to optimize the prediction model.

Main Results:

• The developed approach, ATPBR, achieved a high accuracy of 87.53% and a Matthew's correlation coefficient of 0.554.
• ATPBR demonstrated significantly improved performance compared to existing methods.
• The PSSMPP feature effectively distinguished between ATP-binding and non-binding residues.

Conclusions:

• ATPBR provides a more accurate method for predicting ATP-binding residues.
• The PSSMPP feature and mRMR-IFS selection contribute to improved prediction performance.
• The findings offer insights into the mechanisms of ATP-protein interactions.