Drug-target interaction prediction from PSSM based evolutionary information.

Area of Science:

• Computational biology
• Bioinformatics
• Drug discovery

Background:

• Experimental drug-target interaction prediction is costly and time-consuming.
• In silico methods offer a cost-effective alternative for predicting drug-target interactions.
• Existing computational approaches are broadly categorized into similarity-based and feature-based methods.

Purpose of the Study:

• To develop and evaluate a novel computational model for predicting drug-target interactions using protein sequence information.
• To assess the effectiveness of bi-gram features derived from Position Specific Scoring Matrix (PSSM) for this prediction task.
• To investigate the influence of different negative selection strategies on prediction performance.

Main Methods:

• Extraction of bi-gram features from Position Specific Scoring Matrix (PSSM) of proteins.
• Development of the Bigram-PSSM computational model for drug-target interaction prediction.
• Evaluation of prediction performance using various metrics and comparison of random versus balanced sampling strategies for negative data selection.

Main Results:

• The Bigram-PSSM model demonstrates high-confidence prediction capabilities, particularly for enzyme and ion channel targets.
• The choice of negative selection strategy significantly impacts the performance of drug-target interaction prediction models.
• Balanced sampling generally leads to different performance outcomes compared to random sampling across datasets.

Conclusions:

• The Bigram-PSSM model is a promising feature-based approach for in silico drug-target interaction prediction.
• Careful consideration of negative data sampling is crucial for optimizing computational prediction models.
• This work contributes to advancing computational drug discovery by providing a robust prediction framework.