Human pharmacokinetic simulation from in vitro ADME data and structural descriptors:
Predict pharmacokinetics for oral (po), intravenous (iv) bolus, or infusion regimens, from the integration of ADME data and structural information
Plasma concentration-time profiles and summary PK parameters provided for maximum utility
No specialised staff required
No software licencing costs
Predictions within one working day of receipt of ADME data, if using Cyprotex in vitro ADME data package
Scales seamlessly from a single compound to hundreds of compounds, with no added delay
A rapid screening service for integrating in vitro ADME data and structural information.
How does it work?
Intrinsic clearance measured in either human liver microsomes or human hepatocytes
Fraction unbound in human plasma
Compound structure – as either SDF or SMILES file
Optional human blood to plasma ratio
Optional dose regimen
Predictions are delivered in an Excel workbook containing the following worksheet tabs:
A summary sheet containing summary PK parameters for all combinations of compound and dose regimen simulated
A single sheet for each compound/dose combination, containing:
Plasma concentration-time profile data for 24 hours following administration of final dose
Linear plot of plasma concentration-time profile
Semi-logarithmic plot of plasma concentration-time profile
Predicted PK parameters and plasma profiles can be easily read into databases or modelling software for further manipulation. Plots can be immediately copied into presentations or documents, or formatted to suit specific presentation requirements.
Send pre-generated data along with compound structures. Predictions will be returned within five working days of receipt of the data and structure package.
The models underlying the PK prediction service have been carefully designed and extensively validated in order to provide flexibility of use, and reliability of output.
Plasma concentration-time profiles
The primary output of the service is a predicted plasma concentration-time profile for 24 hours following the end of dosing. These predicted profiles can be used directly for driving PK/PD predictions. The reliability of the predictions has been assessed using test sets of compounds with diverse PK properties. The results are shown in Table 1 for both iv and po administration. The average discrepancy between predicted and measured plasma concentration is approximately 2.5- to 3-fold, except for the prediction of oral dose using microsomes.
Table 1 Statistics for mean-fold error in prediction of plasma concentration following intravenous and oral dosing.
Table shows the mean mean-fold error in the prediction of plasma concentration for up to 24h across diverse test set compounds.
Plasma PK parameters
PK parameters are calculated from the predicted plasma profiles using standard methods. These predicted PK parameters can be used for prioritising compounds for progression to further screens. The reliability of the predictions has been assessed using the same test sets as used for the plasma profiles. The results are shown in Table 2. Clearance, half-life and volume of distribution are all predicted to within 2- to 3-fold, and show high rank-ordering of values, whether using hepatocytes or microsomes. Cmax, AUC and tmax for oral dose administration are all predicted within 2-fold when using hepatocytes, and within 3-fold when using microsomes
Steady state volume of distribution
Elimination phase volume of distribution
Table 2 Statistics for prediction of plasma PK parameters.
MFE = mean-fold error between the predicted and measured PK averaged across all test compounds. *Cmax and AUC are dose- and body-weight normalised in order to calculate the Spearman’s rank correlation coefficient correctly.
Rank = Spearman’s rank correlation coefficient between the predicted and measured PK parameter for the test set compounds.
For illustration, comparisons of predicted and observed Cmax and AUC are plotted in Figure 1.
Figure 1 Comparison of predicted and observed Cmax (top) and AUC (bottom) for oral dose administration.
Predicted parameters were obtained from Cyprotex’s PK prediction service using clearance measured in human hepatocytes. Observed parameters were calculated from clinical plasma profiles. Cmax and AUC are dose- and body-weight normalised.
Questions and answers on PK prediction using ADME data
Please provide an overview of Cyprotex's human PK prediction service.
Cyprotex’s human PK prediction service is a combined in vitro/in silico offering that has been developed to provide reliable prediction of human pharmacokinetics from a minimal set of tier 1 ADME data, making it suitable for use in lead identification (LI) and lead optimisation (LO) stages of drug discovery. At its core is a proprietary in silico model that requires several descriptors calculated from a compound’s structure, as well as measurements or predictions of hepatic metabolic intrinsic clearance (CLint) and plasma protein binding (fup), in order to make compound-specific predictions of human PK for intravenous or oral dose regimens.
How has the model been developed?
The model used in the PK prediction service is a physiologically based pharmacokinetic (PBPK) model. It has been developed using Cyprotex proprietary methodology, and provides unique, Cyprotex-specific predictions. The model has been trained by optimisation against large sets of human in vivo plasma profiles. The optimisation process generates datasets from which models for key compound-specific parameters were then developed by machine learning1 and incorporated into the overall PBPK model framework. This process of optimization followed by machine learning generates robust and interpretable models that provide, in turn, reliable PK predictions.
How has the model been validated?
Sets of compounds having diverse pharmacokinetic characteristics, distinct from the training sets, were selected for testing the model for prediction of intravenous and oral PK. Selection of a diverse set in this manner is necessary in order to obtain an unbiased estimate of the performance of the model.
What is a PBPK model?
A PBPK model is a type of mathematical model that is designed to predict the PK, in plasma, blood and/or other tissues and organs, of a compound after administration to a person or animal2. A PBPK model is a form of simulation model, so named because the goal is to simulate the processes that occur in the time following some specified triggering event. In the specific case of a PBPK model, the triggering event is usually the start of a regimen of one or more drugs, and the processes that are simulated are the distribution, metabolism and elimination of the drug(s), as well as gastrointestinal absorption for an oral dose. The processes are simulated over time following start of the dosing, which enables the familiar concentration-time profiles of drugs to be replicated.
The structures of simulation models can be infinitely variable and, as examples of these, structures of PBPK models that have been developed are also highly variable. The defining characteristics common to most PBPK models are:
The model is conceptualised as a set of compartments. Thus, they are similar to compartmental models routinely used in PK and PK/PD studies. The major distinguishing characteristic is that PBPK models usually have more compartments, each compartment frequently equating to a major tissue or organ in the body. The compartments in the model are then assigned volumes and blood flows, whose values are obtained from knowledge of the physiology of the particular species being modelled. The major organs/tissues represented in the PBPK model that performs our PK predictions are adipose, bone, brain, gut wall, kidney, heart, liver, lung, muscle and skin.
There are one or more additional compartments representing blood in different parts of the circulation (e.g. arteries, veins). These are also assigned volumes appropriate to the particular organism, and a flow through them equal to the cardiac output of that organism. Simulation of an intravenous bolus dose is achieved by setting the amount in the venous compartment equal to the dose amount. The PBPK model is so constructed that the simulated blood flow carries the drug into the other compartments, thus simulating the process of distribution.
Metabolism and other elimination routes can be simulated in a PBPK model. Our model simulates both hepatic and renal clearance routes.
Intestinal absorption of oral dose is frequently modelled as a series of processes occurring in the lumen of the gastrointestinal (GI) tract. These processes can typically include release and dissolution of solid dosage forms, passage of dissolved and solid material along the GI tract lumen, and permeation into the GI tract wall. From here drug passes into the hepatic portal vein, and ultimately to the systemic circulation.
Compounds have compound-specific rates and extents of distribution into the different tissue/organs represented in the model, and different rates of hepatic and renal clearance. The values of these compound-specific parameters are calculated from the inputs to the service (i.e. hepatic CLint, fup and structural descriptors), and thus lead to compound-specific prediction of PK.
What are the benefits of using Cyprotex’s human PK prediction service?
Accuracy and reliability: Cyprotex’s human PK prediction service has unrivalled accuracy and precision in the prediction of intravenous dose PK from simple early ADME screens. Using data from only two tier 1 ADME assays, highly informative PK predictions can be made, permitting more informed compound selection, and facilitating integrated drug discovery.
Self-consistency: An advantage of PBPK models over many other PK prediction tools is that the set of predictions (e.g. all the PK parameters) are qualitatively and quantitatively self-consistent, which is not the case when using multiple models that predict each parameter individually. For example, the clearance, elimination phase volume of distribution and half-life are self-consistent because they are calculated from a single underlying predicted plasma profile. Changes to the properties of a compound will have quantitatively consistent effects on these properties, helping to guide rational drug design. Thus, changing the plasma protein binding of a compound will have quantitatively self-consistent effects on predicted plasma concentration, clearance, distribution and half-life.
Versatility: Cyprotex’s PK prediction service is one of few PK predictors capable of generating reliable predictions for drug PK parameters and plasma concentration profiles from a limited set of early ADME data. Consequently, it provides actionable information of two kinds:
Clients can prioritise compound selection on the basis of predicted PK parameters, e.g. clearance. Having predicted parameters that are both precise and show a demonstrably high level of ranking for all key PK parameters enables selection to be made with a high degree of confidence from the earliest stages of discovery.
For clients keen to initialise quantitative PKPD analysis or modelling as soon as possible, having plasma concentrations predicted within approximately 3-fold of clinical mean values enables very early integration of ADME and activity data to predict potential in vivo activities and/or toxicity liabilities.
Throughput and turnaround: As a consequence of our automated workflow we can return results in a short period of time, independent of whether predictions have been requested for one compound or for hundreds of compounds. Clients who request predictions as part of the combined in vitro-in silico package will receive prediction results within one working day of receiving the requisite in vitro data. Clients providing their own ADME data will receive predictions within 5 working days of receipt of the structures and data package.
1Krstajic D et al., (2014) Cross-validation pitfalls when selecting and assessing regression and classification models. J Cheminform6; 10 2 Brightman FA et al., (2006) Application of a generic physiologically based pharmacokinetic model to the estimation of xenobiotic levels in human plasma. Drug Metab Dispos34; 94-101