Computational models to assign biopharmaceutics drug disposition classification from molecular structure

Area of Science:

• Computational chemistry
• Pharmacokinetics
• Machine learning in drug discovery

Background:

• The Biopharmaceutics Drug Disposition Classification System (BDDCS) is crucial for predicting drug absorption and metabolism.
• Accurate classification of drugs within the BDDCS framework is essential for efficient drug development.

Purpose of the Study:

• To develop and validate in silico models for automatic drug classification according to the BDDCS.
• To assess the performance of machine learning algorithms in predicting BDDCS classes.

Main Methods:

• Machine learning algorithms including recursive partitioning (RP), random forest (RF), and support vector machine (SVM) were employed.
• Molecular descriptors such as clogP, polar surface area, and others were utilized with a dataset of 221 molecules (165 training, 56 test).

Main Results:

• RF model 3 achieved 73.1% accuracy for class 1, RP model 1 achieved 63.6% for class 2, and SVM model 1 achieved 78.6% for class 3.
• Both RP and SVM models demonstrated utility for class 4 predictions.
• Consensus analysis enhanced prediction accuracy for class 2 and 4 drugs.

Conclusions:

• Developed in silico models can predict BDDCS class from molecular structure, utilizing accessible descriptors and software.
• These computational approaches can assist the pharmaceutical industry in accelerating drug development and reducing costs.
• The models offer significant potential in drug discovery for identifying molecules with potential developability challenges early on.