Combinatorial QSAR modeling of human intestinal absorption.

Area of Science:

• Pharmacokinetics and Drug Discovery
• Computational Chemistry
• Machine Learning in Pharmacology

Background:

• Human intestinal drug absorption is critical for drug discovery and development.
• Accurate prediction of drug absorption impacts therapeutic efficacy and safety.
• Existing models often lack comprehensive chemical space representation.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting human intestinal drug absorption.
• To classify compounds as well-absorbed or poorly absorbed.
• To identify key molecular descriptors influencing oral absorption.

Main Methods:

• Utilized a dataset of 458 FDA-approved drugs.
• Calculated 1D-3D physicochemical descriptors.
• Employed feature selection techniques including backward selection, correlation, and decision tree analysis.
• Applied various machine learning algorithms: DTI, LAZAR, SVM, MLP, Random Forests, k-NN, Naïve Bayes.

Main Results:

• Decision Tree Induction (DTI) with CHAID algorithm achieved the best classification rate (88% corrected) and MCC (75%).
• Multilayer Perceptron (MLP) showed the best performance in numeric prediction (RMSE 25.823, R-squared 0.6).
• Identified lipophilic partition coefficients (log P) and hydrogen bonding as important, alongside novel insights into gravitational indices and moments of inertia.

Conclusions:

• Machine learning models can effectively predict human intestinal drug absorption.
• DTI and MLP offer robust approaches for classification and numerical prediction, respectively.
• The study emphasizes the role of structural symmetry and provides generalized models for drug discovery and lead optimization.