An efficient computational method for predicting drug-target interactions using weighted extreme learning machine and...

Area of Science:

• Bioinformatics
• Computational Drug Discovery
• Cheminformatics

Background:

• Predicting novel drug-target interactions (DTIs) is crucial for identifying new drug candidates and therapeutic targets.
• Experimental methods for DTI prediction are time-consuming and costly.
• Developing efficient computational approaches for accurate DTI prediction remains a significant challenge.

Purpose of the Study:

• To propose a novel computational method, WELM-SURF, for identifying drug-target interactions (DTIs).
• To leverage drug fingerprints and protein evolutionary information for enhanced DTI prediction accuracy.
• To provide an efficient computational tool for bioinformatics studies related to DTIs.

Main Methods:

• Utilized Position Specific Scoring Matrix (PSSM) for protein evolutionary information and Speed up robust features (SURF) for sequence feature extraction.
• Employed molecular substructure fingerprints to represent drug chemical structures as feature vectors.
• Applied Weighted Extreme Learning Machine (WELM) classifier for DTI prediction, capitalizing on its fast training and generalization capabilities.

Main Results:

• WELM-SURF achieved high average accuracies: 93.54% (enzyme), 90.58% (ion channel), 85.43% (GPCRs), and 77.45% (nuclear receptor).
• Performance was validated using fivefold cross-validation on diverse biological datasets.
• WELM-SURF significantly outperformed existing methods, including Extreme Learning Machine (ELM) and Support Vector Machine (SVM).

Conclusions:

• The WELM-SURF model demonstrates high accuracy and robustness in predicting DTIs.
• This computational method serves as a valuable tool for advancing bioinformatics research in DTI prediction.
• The findings support the utility of WELM-SURF in accelerating drug discovery pipelines.