Multiclass Classifier for P-Glycoprotein Substrates, Inhibitors, and Non-Active Compounds

Area of Science:

• Pharmacology
• Computational Chemistry
• Toxicology

Background:

• P-glycoprotein (P-gp) is a key transmembrane efflux pump influencing drug absorption, distribution, metabolism, excretion, and toxicity (ADMET).
• Understanding P-gp interactions is crucial for drug discovery and toxicological assessments to prevent intracellular accumulation and reduce drug toxicity.
• Identifying compounds that interact with P-gp is essential for developing safer and more effective drug candidates.

Purpose of the Study:

• To develop a robust in silico multiclass classification model for predicting the probability of a compound interacting with P-glycoprotein (P-gp).
• To utilize a counter-propagation artificial neural network (CP ANN) and 2D molecular descriptors for P-gp ligand prediction.
• To provide a reliable virtual screening tool for early-stage identification of potential P-gp ligands in drug development.

Main Methods:

• Development of a multiclass classification model using a counter-propagation artificial neural network (CP ANN).
• Training and testing the model on an extensive dataset of 2512 compounds, categorized as P-gp inhibitors, substrates, or non-active compounds.
• Validation of the model's performance using an independent external validation set of 385 compounds.

Main Results:

• The in silico model demonstrated strong classification performance with high non-error rate (NER) and average precision (AvPr) values on training and test sets.
• The model achieved NER and AvPr values of 0.93 and 0.93 on the training set, and 0.85 and 0.87 on the test set, respectively.
• External validation yielded NER and AvPr values of 0.70, indicating good generalizability for predicting P-gp ligand potential.

Conclusions:

• The developed in silico classifier is an effective tool for virtual screening of potential P-gp ligands.
• This computational approach can significantly aid in the early stages of drug discovery and toxicological evaluation.
• The model contributes to predicting drug ADMET properties by assessing P-gp interaction probabilities, thereby enhancing drug safety.