Intestinal permeability of drugs together with solubility are the two key factors affecting their oral bioavailability. Absorption module provides accurate predictions of passive intestinal permeability (on jejunal and Caco-2 scales) and extent of oral absorption (%HIA) of drug candidates. These predictions based on intuitive, easily interpretable physicochemical models enable the researchers to rank and select lead compounds according to their permeability across intestinal barrier and to exclude candidates exhibiting extremely poor absorption at the earliest stages.
Caco-2 module contains a mechanistic predictive model of compound permeability across Caco-2 cell monolayer at different experimental conditions. The predictive algorithm was developed using data from >1400 experiments. The resulting model is ionization-specific and also takes into account permeabilities via transcellular and paracellular routes. It uses essential physicochemical properties such as lipophilicity, ionization constants, number of H-Bond donors and molecular size (calculated or experimental if available) as inputs.
Features
Calculates the extent of passive permeability of analyzed compounds across Caco-2 cell monolayers at specified pH and stirring conditions using physicochemical property values lipophilicity and ionization as inputs.
Gives an estimate of relative contributions of different transport routes to Caco-2 permeability.
Allows entering experimentally measured physicochemical properties instead of automatically calculated values to improve the quality of predictions. The experimental conditions (pH and stirring rate) may also be adjusted manually for full flexibility of the simulation.
Support for user-defined input parameters also allows the researcher to model the limiting effect of physicochemical properties on intestinal permeation rate, thus providing a straightforward route for property-based design of oral drugs.
Provides two different modes of expressing lipophilicity and ionization parameters: as LogD at simulated pH point, or as LogP of neutral form together with acid and base pKa values.
Displays the experimental values of the relevant properties for up to 3 similar structures from the built-in database along with each Caco-2 permeability prediction.
Interface
Main physicochemical descriptors that are used as inputs for calculation. By default, automatically calculated values are used. Click and type a new value to model the limiting effect of lipophilicity, hydrogen bonding acidity and molecular size on Caco-2 permeability
Select which type of lipophilicity descriptor - LogP or LogD - to use in calculations.
Calculated acid and base pKa. When In LogP mode, click to select or type new values to model the limiting effect of ionization constants on Caco-2 permeability. In LogD mode, these dropdowns are read-only and only serve to display calculated pKa for informational purposes. In this mode, pKa values are not directly used in calculations, as the effect of ionization is already incorporated in LogD.
a. Click to revert to an automatically calculated property value (LogP in this picture) for a compound and to recalculate the Caco-2 permeability
b. Click to recalculate the Caco-2 permeability using the currently specified parameter values
Experimental conditions used in simulation. Type in new values to change the pH or stirring rate (rpm)
Estimated value of Caco-2 permeability (Pe, cm/s) and the relative contributions of transcellular and paracellular pathways, calculated as a function of compound structure and provided parameters
Up to 3 most similar structures in the built-in database with experimental values and references
Technical information
Only a short summary of main technical aspects of Human Intestinal Absorption predictor is given here. For a more detailed description of the modeling approach and underlying theory please refer to J Pharm Sci.2009;98(11):4039-54. [1]
Caco-2 permeability predictor was developed using the same general principles. However, this model has been recently re-parameterized using a larger data set and a simplified approach based on a composite LogD descriptor instead of separate LogP and pKa parameters. More information about the new Caco-2 permeability model can be found in J Pharm Sci.2019;108(1):78-86. [2]
Calculated quantitative parameters
The main output of the Passive Absorption module is the maximum achievable extent of human intestinal absorption (when solubility is not a limiting factor) expressed as a percentage value and denoted %HIA. Typically, compounds exhibiting %HIA > 70% are considered well absorbed, those with %HIA < 30% - poorly absorbed, while values in the range 30 - 70% represent moderate absorption. Intestinal permeability rates corresponding to the respective %HIA values are calibrated on several scales and given as:
Effective jejunal permeability coefficients at pH 6.5 (Pe, 10-4 cm/s)
Absorption rate constants (ka, min-1).
Additionally, Passive Absorption\Caco-2 module presents effective permeability coefficients in Caco-2 monolayers (Pe, 10-6 cm/s) at user-defined conditions (pH and stirring rate).
Descriptors & Modeling Method
Descriptors used for modeling included key physicochemical properties - octanol/water LogP of neutral species as a determinant of lipophilicity, ion form fractions at pH 6.5 calculated from the respective pKa values, number of hydrogen bond donors in the molecule, and McGowan characteristic volume reflecting molecular size. All physicochemical parameter values were calculated with Algorithm Builder software (Quant Struct Act Relat.2002;21(1):23-37. [3])
Due to the evidence of sigmoidal relationship between fraction absorbed (represented by experimental data) and actual permeation rates data were modeled using non-linear least squares regression. Fitting was performed in a multi-step approach with separate stages employed for:
Determination of paracellular transport parameters
Describing trancellular diffusion of non-electrolytes
Estimating ionization dependence of intestinal absorption
The sigmoid relationships between %HIA and octanol/water log P for various electrolyte classes are illustrated in the figure below. The sigmoids obtained for charged species are shifted relatively to the respective curve for non-electrolytes, although the shift is not as marked as could be expected if absorption was modeled by pH-dependent octanol/water distribution coefficient log D. Ionization-specific analysis was therefore performed to devise more precise corrections to log P values needed for prediction of passive jejunal permeability.
Non-linear dependence of fraction absorbed on lipophilicity for compounds representing different ionization states. Only compounds with MW > 250 were considered to minimize the contribution from paracellular absorption route). Bases: pKaB > 7, acids: pKaA < 5.
In case of Caco-2 permeability, analysis of the new expanded data set including many representatives of novel pharmaceutical classes revealed even less pronounced differences to octanol/water system in terms of the effect of ionization. Accordingly, no significant differences in prediction accuracy were observed between models based on separate log P/ion form fraction descriptors, or a single log D descriptor accounting for both lipophilicity and ionization influence. Therefore, the latter approach was preferred due to simplicity, and considering the fact that in many recent publications lipophilicity of novel chemicals is determined as log D at relevant pH. Nevertheless, the current implementation of Caco-2 permeability predictor in ACD/Percepta allows for a more flexible representation of physicochemical properties and accepts either log D, or log P + pKa values as inputs for calculation. In the latter case log P is recalculated to log D, which is then supplied to Caco-2 model.
Prediction accuracy
The final quantitative %HIA data set (cleaned of values distorted by P-gp efflux, facilitated diffusion and other side processes) that was used for modeling consisted of 567 %HIA values, mostly for marketed drugs or drug candidates. For validation purposes two independent test sets were compiled from several publications dealing with jejunal absorption of drugs. External validation sets represented other types of experimental data that were not used in model development:
The first set contained directly measured jejunal permeability coefficients (Peff) for 25 compounds extracted from Xenobiotica.2007;37(10-11):1015-51. [4]
The second set was comprised of absorption rate constants (ka) for 22 molecules from J Med Chem.2006;49(12):3674-81. [5].
The main advantage of such type of validation is the possibility to evaluate the intrinsic correctness of our model rather than just goodness of fit between experimental and predicted HIA.
Model performance on internal %HIA data set and external validation sets is summarized in the table below:
Data set
N
R2
RMSE
%HIA training set
567
0.93
9.5%
log Peff validation set
25
0.72
0.45
log Ka validation set
22
0.84
0.35
Note that RMSE for the training set is grayed since prediction error for percentage values (that are mostly concentrated on the ends of the scale) does not provide an unambiguous measure of model quality, while the actual accuracy of predictions is best illustrated by model performance on external validation sets yielding RMSE about 0.4 log units which is close to the error of experimental determination.
The respective results for Caco-2 permeability model are as follows:
Data set
N
R2
RMSE
Training set
497
0.80
0.44
Internal validation set
442
0.77
0.49
External validation set
427
0.53
0.47
A lower R2 value was obtained for external validation set because this set had a much narrower variation
range of both log D, and log Pe values. Nevertheless, consistently low RMSE values indicate that the model is not-overfitted and can predict passive permeability close to the level of experimental uncertainty.
Reference database
In addition to HIA and Caco-2 permeability predictions, ACD/Percepta provides Absorption DB - a a browsable database of human intestinal absorption & bioavailability. The experimental data were compiled from reference pharmacokinetic tabulations and original articles, the main sources being "Therapeutic Drugs" (ed. by C. Dollery), Goodman & Gilman's "The Pharmacological Basis of Therapeutics", and a compilation by Zhao, Y. H. et al. J Pharm Sci.2001;90(6):749-84. [6] A qualitative assignment of absorption category (good, moderate, poor) is provided for each compound in the database along with comments regarding the quantitative extent of absorption/bioavailability, presence of carrier-mediated transport, etc. where available.
Transport mechanisms
The predictive models comprising the Passive Absorption module account for passive transport of analyzed compounds across intestinal barrier. In order to propose a clear physicochemical explanation of passive diffusion process, the data used for model development were thoroughly evaluated to exclude values affected by enzymatic efflux or influx. For such compounds special alerts are displayed in the Oral Bioavailability module indicating that expected oral bioavailability/absorption values for these molecules may be higher or lower due to presence of carrier-mediated processes.
