DDR: efficient computational method to predict drug-target interactions using graph mining and machine learning...

Area of Science:

• Computational biology
• Bioinformatics
• Drug discovery

Background:

• Computational drug-target interaction (DTI) prediction offers a cost-effective strategy for identifying new DTIs.
• Existing DTI prediction methods often exhibit high false positive rates, limiting their clinical utility.

Purpose of the Study:

• To develop a novel computational method, DDR, for improving the accuracy of drug-target interaction prediction.
• To address the limitations of current DTI prediction techniques, particularly their high false positive rates.

Main Methods:

• Development of DDR, a method employing a heterogeneous graph integrating known DTIs and multiple drug/protein similarities.
• Application of a non-linear similarity fusion technique with a heuristic pre-processing step for optimized similarity combination.
• Utilization of a random forest model with graph-based features extracted from the heterogeneous graph.

Main Results:

• DDR significantly reduces prediction error across various scenarios: 34% for novel drugs, 23% for novel targets, and 34% for known drugs/targets with unknown interactions.
• Validation using independent evidence confirmed 22 out of the top 25 novel predictions made by DDR.
• The method demonstrates a substantial improvement in accuracy compared to state-of-the-art DTI prediction approaches.

Conclusions:

• DDR presents a significant advancement in computational drug-target interaction prediction.
• The method's ability to accurately identify novel DTIs suggests its potential as an efficient tool in drug discovery pipelines.
• The provided data and code facilitate reproducibility and further development in the field.