Permeability across monolayers of the Caco-2 line of human epithelial colorectal adenocarcinoma cells is a common in vitro model used to assess potential of compounds to be absorbed across the human intestine. The apparent permeability (Papp) from the apical to basolateral has been shown to correlate with the in vivo fraction absorbed following oral administration of a drug (Artursson P. and Karlsson J., Biochem. Biophys. Res. Comm.175(3), pp. 880–5, 1991). A criterion commonly applied to select compounds with a higher chance of achieving good oral absorption is a Papp value of 1×10-5 cm/s or higher.
Nordqvist et al. (QSAR & Comb. Sci. 23(5), pp. 303–310, 2004) published experimental Caco-2 Papp values for a set of small molecules, including marketed drugs. We have applied the StarDrop’s Auto-Modeller™ to these data to generate models to predict log(Papp). Two data sets were published with this paper, a training set containing 77 compounds and an independent test set of 23 compounds. For consistency, we have used the same sets to train and validate the models generated.
The training set is smaller than we would ideally like for building a global model of a complex property. To mitigate this, we have used only ten general descriptors to build the models, in an effort to avoid over training. The descriptors used were: logP, McGowan’s volume, number of hydrogen bond acceptors and donors, flexibility, topological polar surface area, number of aromatic rings, overall charge, negative charge and positive charge. Please see the detailed model output (which can be downloaded below) and the StarDrop Reference Guide for detailed definitions of these descriptors.
All of the modelling techniques available in the Auto-Modeller were applied and the best models resulted from the PLS and Gaussian Processes (Forward Variable Selection) methods. The results are very comparable to those published in the paper by Nordqvist et al. and are summarised in the table below:
| Modelling method | Training set | Test set | ||||
|---|---|---|---|---|---|---|
| R2 | r2corr | RMSE | R2 | r2corr | RMSE | |
| PLS | 0.47 | 0.47 | 0.54 | 0.66 | 0.72 | 0.50 |
| GPFVS | 0.60 | 0.60 | 0.47 | 0.66 | 0.70 | 0.50 |
The data sets and detailed outputs from the modelling process may be downloaded.
To use these within StarDrop, download and save these files in a convenient place.
Load them into StarDrop using the folder button on the Models tab.
Alternatively, the directory in which the model files have been saved can be added to the paths from which models are automatically loaded when StarDrop starts by selecting the File->Preference menu option and adding the directory under Models in the File Locations tab.
The data sets and detailed outputs from the modelling process may be downloaded.
To use these within StarDrop, download and save these files in a convenient place.
Load them into StarDrop using the folder button on the Models tab.
Alternatively, the directory in which the model files have been saved can be added to the paths from which models are automatically loaded when StarDrop starts by selecting the File->Preference menu option and adding the directory under Models in the File Locations tab.
Ann and Ajay discuss the science behind and applications of the eSim molecular similarity method, a ligand-based drug design approach which considers surface-shape, electrostatics, and directionally sensitive hydrogen-bonding when comparing two molecules.
This article is a collaboration with Intellegens, the University of Cambridge and AstraZeneca. It provides a proof-of-concept study in which Cerella™ is used to predict rat in vivo pharmacokinetic (PK) parameters and concentration–time PK profiles.
