I’m working on brain modeling and penetration prediction, and I would like to use Dr Shah’s model. It’s a full PBPK model with a brain compartment divided in different sub-spaces. So I code it on Phoenix. In a first step, I wanted to validate the human version of the model with plasma data. So based on a reference described in the study I created the dataset for the model.
I have 2 questions:
The profile changes based on #Points used in the simulation. For the same X range (Max X), comparing 100 vs 1000 #points, the simulated points (DV) are different for a same time (IVAR). From my understanding of Phoenix software, I expected to find the same value for a same time point. Could you explain me why we observed different simulated point for a same time point? What does the term “#points” mean exactly?
Similarly, the difference between profiles from “Ind DV,IPRED vs IVAR” plot and “Ind Simulation” plot is not clear in terms of modeling. Even if the predictions in the first plot are not well, it follows the trend of the observations. Whereas in the “Ind Simulation” plot, the profile is fully different (when #points = 1000). From my understanding, this should lead to close results.
The project file is enclosed to this post in order to allow you to a look into it. The model is quite big but not that complex.
re 1. you should expect almost identical values for the two simulatons. There was just a problem with the ODE solver in both of your simulation objects. If you inspect the Warning and Error output, you could see that:
re 2. The observations and predictions are several orders of magnitude different. I can only guess that some unit conversions are missing. Since I don’t see how you got to the estimates for your simulations it is difficult to tell.
In addition, for your information: the #points lets you decide on the x-resolution of your simulated data, it means the number of points to simulate. For your first simulation you chose 100 points over and x-range of 1000, that means calculate values for 0,10,20,… For your second simulation you chose 1000 points over the same range, e.g. values for 0,1,2,3,…
Thanks a lot for your reply and your solution! Indeed, switch ODE mode to auto-detect fix the problem. May I ask you what the difference between these modes is?
For the R2, with your proposed solution, both plots (“Ind DV,IPRED vs IVAR” and “Ind Simulation”) are much more consistent even if the prediction are still not good. The estimates for the simulations came from an article I’m trying to reproduce the model (Chang et al. 2019).
ref : Chang, Hsueh-Yuan, Shengjia Wu, Guy Meno-Tetang, and Dhaval K. Shah. 2019. “A Translational Platform PBPK Model for Antibody Disposition in the Brain.” Journal of Pharmacokinetics and Pharmacodynamics, May. https://doi.org/10.1007/s10928-019-09641-8.
There is a lot of background information to this. Essentially it comes down to the question of how the software shall solve differential equations. Phoenix gives the user a choice among several methods. For an overview please see our online help:
In simple practical terms, whenever you deal with typical PKPD models and linear kinetics use the default option (matrix exponent). When you have huge differences in coefficients or rate constants switch to auto-detect. Same holds for high-non linearity systems.
