Classification of drug molecules considering their IC50 values using mixed-integer linear programming based...

Area of Science:

• Computational chemistry
• Cheminformatics
• Drug discovery

Background:

• Computational approaches aid in identifying drug candidates by predicting activity on target proteins.
• Existing methods for drug activity prediction are numerous, but improvements in classification accuracy are sought.

Purpose of the Study:

• To enhance drug activity classification accuracy using a novel algorithm.
• To identify significant molecular descriptors influencing drug activity.
• To classify drug molecules as low or high active based on binding affinity (IC50 values).

Main Methods:

• Developed a classification algorithm combining partial least squares regression with mixed-integer programming based hyper-boxes.
• Applied the method to known inhibitor datasets (ACHE, BZR, DHFR, COX-2) and a custom dataset for Cytochrome P450 C17 inhibitors.
• Compared performance against 63 other classification methods, including SVM and Naïve Bayes.

Main Results:

• Achieved high classification accuracies, with a worst-case prediction accuracy of 96% for known inhibitor datasets.
• Attained 100% accuracy in predicting activities for Cytochrome P450 C17 inhibitors.
• Demonstrated superior performance compared to 63 existing classification methods.

Conclusions:

• The developed approach effectively predicts inhibitory effects of compounds based on molecular descriptors.
• This method enhances the drug discovery process by saving time and resources.
• The approach offers a valuable tool for predicting drug efficacy and optimizing candidate selection.