Drug-target interaction prediction via multiple classification strategies.

Area of Science:

• Computational biology
• Pharmacology
• Bioinformatics

Background:

• Drug-target interaction (DTI) prediction is crucial for drug discovery but faces challenges due to imbalanced interaction data across protein targets.
• Existing methods struggle with targets having few positive interaction samples, leading to suboptimal prediction performance.

Purpose of the Study:

• To develop a novel drug-target interaction prediction method, Multiple Classification Strategies for DTI (MCSDTI), that addresses the issue of imbalanced interaction data.
• To improve the accuracy and effectiveness of DTI prediction by employing tailored classification strategies for different target groups.

Main Methods:

• MCSDTI categorizes protein targets into two groups: those with a smaller number of interactions (TWSNI) and those with a larger number of interactions (TWLNI).
• Distinct classification strategies are applied to TWSNI and TWLNI to predict drug-target interactions.
• Independent evaluation of TWSNI and TWLNI prevents results from being dominated by targets with abundant interactions.

Main Results:

• MCSDTI demonstrated superior performance across five DTI datasets (NR, IC, GPCR, E, DB).
• The method achieved higher AUCs compared to the second-best methods for both TWLNI (e.g., 3.31% higher on NR) and TWSNI (e.g., 3.20% higher on IC).

Conclusions:

• MCSDTI is a competitive DTI prediction method, outperforming previous approaches on most datasets and target parts.
• Employing diverse classification strategies for different target groups is an effective approach to enhance DTI prediction accuracy.