Comparing multilabel classification methods for provisional biopharmaceutics class prediction.

Area of Science:

• Pharmacokinetics and Drug Development
• Computational Chemistry and Cheminformatics

Background:

• The Biopharmaceutical Classification System (BCS) is crucial for immediate-release oral dosage form development and biowaivers.
• Multilabel classification offers a framework for predicting multiple drug properties simultaneously, unlike traditional single-label methods.

Purpose of the Study:

• To compare the efficacy of two multilabel classification methods, binary relevance and classifier chain, for provisional BCS class prediction.
• To evaluate the potential of these methods for enhancing biological property predictions in drug discovery.

Main Methods:

• Utilized large datasets of drug permeability and solubility from literature to build decision tree models.
• Validated individual permeability and solubility models.
• Compared binary relevance and classifier chain methods using a BCS validation set of 127 compounds.

Main Results:

• The classifier chain method, which accounts for label interdependencies, outperformed the binary relevance method in provisional BCS class prediction.
• Decision tree models were developed for predicting drug solubility and permeability.

Conclusions:

• The classifier chain method demonstrates significant potential for improving the accuracy of biological property predictions.
• This comparative study highlights the advantages of considering label interactions in multilabel classification for biopharmaceutical applications.