I am new to Pumas and modeling - killer combo! I have seen some recent posts on LinkedIn about you, so I thought I could ask for your advice. Not sure if this is the right way to approach this forum. I guess other experts can weigh in as well. I wanted to check if you can confirm is the following model syntax looks right. My model has a one compartment disposition, a slow absorption. Tlag was deemed to inappropriate so I was trying a transit compartment absorption model.
My code begins here
ka_transit2_1cmt = @model begin
tvmtt ∈ RealDomain(lower=5, upper=100.0)
tvcl ∈ RealDomain(lower=0.001)
tvvc ∈ RealDomain(lower=0.001)
σ²_add ∈ RealDomain(lower=0.001)
CL = tvcl
Vc = tvvc
ktr = 1/tvmtt
Depot' = -ktr * Depot_slow
Transit1' = ktr * (Depot_slow - Transit1)
Transit2' = ktr * (Transit1 - Transit2)
Central' = ktr * Transit2 - CL/Vc * Central
cp = @. (Central/Vc)
conc ~ @. Normal(cp, sqrt(σ²_add))
My code ends here
Welcome to the community and we are glad to have you on board!
You can take a look at some of the absorption model examples here Absorption models
Just looking at the code (on my phone), it seems correct what you have coded up there.
Dr Vijay - Thank you very much. May be you or some other expert can help me with understanding the statistical implications of my parameterization - ktr vs mtt. Would there be any preference in adding a between-subject variability on one over the other? Typically kinetic models are specified from first principles in rates and rate-constants; and not in terms of time (except for tlag). Would the community recommend specifying the BSV on ktr; and derive mtt from the reciprocal individual estimate of ktr? Or directly specify the variance on mtt? Hope I am not asking a very silly question. I do not want to take undue advantage of you; so if you think I should write to another expert, please suggest.
Feel free to ask away. This is an open inclusive community for everyone to learn and contribute
Not sure if there is a preference one way or the other.
MTT seems like a parameter that has a better interpretation than
ktr in my opinion. I would add BSV there. However, these are just a play of numerical arrangements at the end of the day. We should use something that helps us articulate our results better.
I absolutely agree with @vijay. One thing you could do is try it both ways (BSV on ktr and BSV on MTT) and see if one of the 2 models seems better (e.g. lower OFV), or the distribution of the random effects looks more normal or one or the other parameter (maybe look at QQ-plots). Just a suggestion.