A Machine Learning Approach for Drug-target Interaction Prediction using Wrapper Feature Selection and Class...

Area of Science:

• Bioinformatics
• Computational Chemistry
• Pharmacology

Background:

• Drug-target interactions (DTIs) are vital for drug discovery, repositioning, and understanding side effects.
• The increasing volume of genomic and drug data complicates the identification of new DTIs.
• Existing datasets for training predictive models often suffer from class imbalance and high dimensionality.

Purpose of the Study:

• To develop a computational model for predicting drug-target interactions (DTIs).
• To address challenges of class imbalance and high dimensionality in DTI prediction datasets.
• To identify novel drug-target associations for Enzyme, Ion Channel, G Protein-Coupled Receptor (GPCR), and Nuclear Receptor protein classes.

Main Methods:

• Utilized dipeptide composition for target protein sequences and molecular descriptors for drugs.
• Employed a machine learning approach incorporating wrapper feature selection and Synthetic Minority Oversampling Technique (SMOTE).
• Evaluated the model on four distinct protein classes using 10-fold cross-validation.

Main Results:

• Achieved high accuracy (up to 95.9%) and precision (up to 96.3%) across the studied protein classes.
• The model successfully predicted new drug-target interactions absent from the training data.
• Identified potentially important features for understanding DTIs within these protein categories.

Conclusions:

• The proposed method effectively predicts drug-target interactions, aiding in the discovery of new therapeutic associations.
• The developed computational model offers a valuable tool for researchers investigating drug-target relationships.
• A standalone package is available for predicting DTIs, facilitating broader application in drug discovery research.