Generative topographic mapping-based classification models and their applicability domain: application to the...

Area of Science:

• Computational chemistry
• Cheminformatics
• Drug discovery

Background:

• Generative topographic mapping (GTM) has been previously shown effective for molecular descriptor data visualization and classification.
• Modeling in the latent space of molecular descriptors is crucial for understanding complex relationships.

Purpose of the Study:

• To apply generative topographic mapping (GTM) for classifying the four classes of the BioPharmaceutics Drug Disposition Classification System (BDDCS).
• To develop and evaluate new definitions for the applicability domain (AD) of predictive models in cheminformatics.
• To assess the utility of different AD definitions for BDDCS modeling and interpretability.

Main Methods:

• Utilized generative topographic mapping (GTM) for modeling in a two-dimensional latent space.
• Employed VolSurf descriptors for the classification of BioPharmaceutics Drug Disposition Classification System (BDDCS) classes.
• Proposed and compared three novel applicability domain (AD) definitions: class-independent (GTM likelihood), class-dependent (predominant class), and class-dependent (informational entropy).

Main Results:

• The class entropy applicability domain (AD) demonstrated the highest efficiency for BDDCS modeling.
• The predominant class AD provides direct visualization on GTM maps, aiding model interpretation.
• GTM successfully models BDDCS classes in a reduced two-dimensional latent space.

Conclusions:

• Generative topographic mapping (GTM) is a powerful tool for BDDCS classification and visualization.
• The class entropy AD is the most effective for ensuring model reliability in this context.
• Visualizable ADs, like the predominant class method, enhance the practical interpretation of predictive models in drug disposition studies.