In silico modelling of chylomicron association to predict lymphatic absorption of small molecules
Jed Malec, Jong Bong Lee, Atheer Zgair, Pavel Gershkovich
School of Pharmacy, University of Nottingham, Nottingham, UK
Objectives: Association of drugs with chylomicrons (CM) in the enterocytes is a key process involved in intestinal lymphatic transport of drugs. The association can be measured by means of previously established ex vivo assay for the prediction of lymphatic transport potentials. The aim of this study was to generate an in silico model for prediction of association of drugs with CM based on physicochemical properties of molecules.
Methods: Ex vivo data for CM association has been obtained from published articles (1-3) or obtained in our laboratory employing methods established in the literature. The modelling was performed using R software version 3.3.2 (4) with packages caret (5) and Random Forest (6). Three data sets have been compared: whole data set (55 compounds), cannabinoids and bexarotene (BEX) prodrugs, and BEX prodrugs only. For each data set, a test set of approximately 20% of compounds of the corresponding set has been defined to assess the accuracy of the models by means of 5-fold cross-validation and calculating root-mean square errors (RMSE). Influence of the descriptors was assessed by means of recursive feature elimination (RFE), which selected the features to be used by RF model, or by the Random Forest’s built-in feature ranking function.
Results: Comparison of most influencing descriptors and test set RMSE for models made using RF alone or using RFE prior to RF algorithm showed that both methods perform equally well. Two new descriptors of distribution coefficient per heavy atoms count (LogD7.4/HA) and polar surface area per molecular volume (PSA/MV) were found to be important for the models. The most influential descriptors were LogD7.4, LogD7.4/HA, PSA/MV, LogP-LogD7.4 and formal charge. After excluding descriptors with little physical value, a final model was established using RF alone and using whole data set. The simulation gave 12% RMSE and r2=0.84, for which the RMSE decreased to 10% after normalisation using the regression line.
Conclusions: A predictive model was developed ensuring high confidence in the values obtained. The model performed well for compounds representing different scaffolds. The accuracy of the model is limited by the size of the data set, the chemical space it covers and the physicochemical properties employed here. Nevertheless, our analysis highlights the importance of lipophilicity and polar interactions for the association of compounds with CM, a process that has not been described in detail yet.
References:
[1] Gershkovich P, Fanous J, Qadri B, et al., The role of molecular physicochemical properties and apolipoproteins in association of drugs with triglyceride-rich lipoproteins: in-silico prediction of uptake by chylomicrons, J Pharm Pharmacol, 61 (2009): 31-39.
[2] Borkar N, Chen Z, Saaby L, et al., Apomorphine and its esters: Differences in Caco-2 cell permeability and chylomicron affinity, Int J Pharm, 509 (2016): 499-506.
[3] Lee JB, Zgair A, Soukarieh F, et al., Targeting bexarotene to the intestinal lymphatic system by lipophilic prodrugs approach, Proceedings of World Meeting of Pharmaceutics and Biopharmaceutics and Pharmaceutical technology, Glasgow, UK, 08 Apr 2016.
[4] R software: R Core Team (2016). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria; http://www.R-project.org/.
[5] Caret package: Max Kuhn. Contributions from Jed Wing, Steve Weston, Andre Williams, Chris Keefer, Allan Engelhardt, Tony Cooper, Zachary Mayer, Brenton Kenkel, the R Core Team, Michael Benesty, Reynald Lescarbeau, Andrew Ziem, Luca Scrucca, Yuan Tang, Can Candan and Tyler Hunt. (2016). caret: Classification and Regression Training. R package version 6.0-73. https://CRAN.R-project.org/package=caret
[6] Random Forest package & algorithm: A. Liaw and M. Wiener (2002). Classification and Regression by randomForest. R News 2(3), 18--22.
