Poster Presentation
Development of a Quantitative Physicochemical Model for Predicting P-glycoprotein Inhibition
Kiril Lanevskij, Development Project Manager, ACD/Labs
K. Lanevskij (1), R. Didžiapetris (1), A. Sazonovas (1)
(1) ACD/Labs, Inc., Toronto, Ontario, Canada
Human P-glycoprotein (P-gp) is a carrier protein of ATP-Binding Cassette (ABC) superfamily that is the major contributor to protection of tissues from a wide variety of xenobiotics and the principal factor behind the phenomenon known as Multi-Drug Resistance (MDR) in cancer treatment. Inhibition of P-gp activity may clinically manifest by MDR reversion and also lead to drug interactions, when a P-gp inhibitor enhances bioavailability of another co-administered drug, which is a P-gp substrate.
The inhibitory effect is usually quantified by the compound’s P-gp half-inhibitory constant IC50. However, most QSAR studies treat P-gp inhibition as a binary endpoint and only produce classification models to distinguish P-gp inhibitors from non-inhibitors at a predefined IC50 threshold value. The situation is further complicated by the fact that many experimental studies do not determine exact IC50s, but only report open intervals indicating that the compound’s IC50 is greater or less than a certain value.
Recently, we have analyzed a P-gp efflux dataset similar in composition and demonstrated that such kind of data is amenable to quantitative modeling using censored regression-based machine learning [1-2]. Moreover, P-gp exhibits extremely broad ligand specificity, meaning that the vast majority of ligands can be identified purely by their physicochemical properties. In the current study, we investigated a similar approach for modeling P-gp inhibition using a dataset of almost 2000 compounds with exact or censored IC50 values collected from literature. According to the preliminary results, an even simpler model trained on a small subset of data with exact values and a minimal set of descriptors, accounting for lipophilicity,
ionization and topology of the molecules, can classify the remaining compounds at a threshold of IC50 = 10 μM with over 85% accuracy.
[1] Lanevskij Kiril, Didžiapetris Remigijus, Sazonovas Andrius, 2023, ‘Physicochemical QSAR Model of P-glycoprotein Efflux Ratio Based on Quantitative and Censored Data’, Toxicol Lett, 384, S118.
[2] Lanevskij Kiril, Didžiapetris Remigijus, Sazonovas Andrius, 2024,‘Physicochemical QSAR model of P-glycoprotein efflux ratio and its application to predicting brain penetration’, Toxicol Lett, 399, S144-S145.