RFDT: A Rotation Forest-based Predictor for Predicting Drug-Target Interactions Using Drug Structure and Protein...

Area of Science:

• Bioinformatics
• Computational Chemistry
• Drug Discovery

Background:

• Identifying drug-target interactions (DTI) is crucial for new drug candidate discovery.
• Experimental DTI identification methods are time-consuming, expensive, and challenging.
• Urgent need for computational methods to predict potential DTIs.

Purpose of the Study:

• Develop a novel computational model for predicting potential drug-target interactions (DTI).
• Utilize structural properties of drugs and evolutionary information of proteins for DTI prediction.

Main Methods:

• Encode protein sequences using Position-Specific Scoring Matrix (PSSM) for evolutionary information.
• Encode drug molecules using fingerprint feature vectors for structural properties.
• Employ the Rotation Forest (RF) model for predictive modeling on benchmark datasets.

Main Results:

• Achieved high prediction accuracies: 91.3% (enzymes), 89.1% (ion channels), 84.1% (GPCRs), and 71.1% (nuclear receptors).
• Outperformed state-of-the-art Support Vector Machine (SVM) classifiers.
• Demonstrated superior performance compared to other existing methods.

Conclusions:

• The proposed computational method is effective for predicting drug-target interactions.
• This approach can significantly assist in new drug research and development.