Jump to content


Photo

QRPEM fatal error


  • Please log in to reply
50 replies to this topic

#21 Elliot Offman

Elliot Offman

    Advanced Member

  • Members
  • PipPipPip
  • 101 posts

Posted 31 March 2013 - 12:47 AM

I'm one of your Canadian clients ;)

our office is in Montreal.

 

In terms of this project, you are correct, I'm modeling a single-dose subQ macromolecule injection. We know that there is lymphatic uptake of proteins however my data clearly suggests a pseudo-zero order input and I've got a very intenstive sampling regimen to characterize this input. This is followed by a more linear increase in slope. This cannot be modeled with simply a first order ka or even 2 first order ka's split between the central and absorption compartment. Hence my proposal to allow a fraction to input directly into the lymph and a fraction via first order into the vasculature.

 

There is some basis for this approach from PBPK models and I am attempting to use the NLME program to characterize this approach.

 

For the point about the covariate, can you elaborate why we would put this covariate in?

when you state "in the dataset" what goes in the dataset? I'm not sure I understand this part. although I can understand why dose itself could be a covariate. Are we simply stating that dose is a covariate so it maps the dose and dose time and knows there is only a single-dose administered?

 

 

Incidentally, and not related to my original question, but I know from your Webex's you are a quite the wizard with these things.

 

What in your opinion is the best way of calculating total Clearance when you have CL from the central compartment and the target mediated clearance which includes a rate of internalization of the receptor-protein complex?

 

Any help here would also be appreciated

 ×ª×•×“×”

Elliot



#22 Elliot Offman

Elliot Offman

    Advanced Member

  • Members
  • PipPipPip
  • 101 posts

Posted 31 March 2013 - 12:16 PM

I just want to confirm that in the covariate column mapped to "dosing" is the same information just copied over, from the dosing columns that get mapped to Alymph and Aa?



#23 Elliot Offman

Elliot Offman

    Advanced Member

  • Members
  • PipPipPip
  • 101 posts

Posted 31 March 2013 - 03:48 PM

So I think its just easier to share my project.

hopefully I've redacted anythign confidential.

 

Anyway, in the project, Model B is the code Serge provided.

Model C, is where I think there were some error which I corrected but otherwise the same.

 

Model A is what I would have done myself without knowing about the covariate step.

 

Neither model is mapping correctly as don't have the ability to select initial estimate, which is usually my clue thiere is an error somewhere, but I can't figure it out.

 

I'm not sure why the code Serge provided doesn't map. But project is attached.

(I'm only modeling treatment B).

Elliot [file name=TMDD_model_with_parallel_absorption.phxproj size=501829]http://www.pharsight.com/extranet/media/kunena/attachments/legacy/files/TMDD_model_with_parallel_absorption.phxproj[/file]

Attached Files



#24 serge guzy

serge guzy

    Advanced Member

  • Members
  • PipPipPip
  • 485 posts

Posted 31 March 2013 - 06:21 PM

Dear Elliot

In the statement

Observe ... it should be

observe

The capital O is not accepted.

By the way, is the talg in the absorption or in the lymphs? Currently it is in the absorption compartment.

best

Serge



#25 serge guzy

serge guzy

    Advanced Member

  • Members
  • PipPipPip
  • 485 posts

Posted 31 March 2013 - 06:25 PM

By the way the code I provided is only for single dose. Is it the case?

best

Serge



#26 serge guzy

serge guzy

    Advanced Member

  • Members
  • PipPipPip
  • 485 posts

Posted 31 March 2013 - 06:53 PM

Dear Elliott

XF is already in the logit domain and therefore should not have any lower and upper bound. XF goes from -infinite to + infinite and F from 0 to 1.

Your current version will have trouble finding standard errors. If you really need it, look how juch time it takes to fit the data set. Once it is done, save the project, accept all fixed an random and then do let us 10 bootstraps (option in the run option). It will give you some sense of how se's are.

If you are using QRPEM, select 3000 as number of samples, 20 iterations for start, click on advanced option and select the MAP.

I am not sure tlag is necessary and think your model may be way overparametrized.

You can look at initial estimates and play with all dmoel parameters to see who are the one that are sensitive and who are the one that are not sensitive.

If you type observe instead of Observe, you should be able to see initial estimate time profiles.

best

serge



#27 Elliot Offman

Elliot Offman

    Advanced Member

  • Members
  • PipPipPip
  • 101 posts

Posted 31 March 2013 - 06:53 PM

I made a typo with that observe capital "O". I need to QC my work better.

 

yes, single-dose SC.

My assumption is that Tlag is in the Aa or 1st-order, since if you look at the data I sent, there is a plateau followed by the1st order absorption.

Elliot



#28 serge guzy

serge guzy

    Advanced Member

  • Members
  • PipPipPip
  • 485 posts

Posted 31 March 2013 - 06:57 PM

Typo error occur, not a big deal. now if you want QRPEM, do 3000 samples, 20 iterations and click on advanced options and select MAP, no standard errors.

If you want some se assessment, do a bootstrap after you get the results.

You copy the model to the workflow, then accept all fixed and random, select bootstrap from run options, 10 bootstraps just to get some results as each bootstraps will take a while.

In the upcoming version, you will be bale to get vey good se estimates in a relatively short timeframe.

Are you sure this model is not overparametrized?

Best

Serge



#29 Elliot Offman

Elliot Offman

    Advanced Member

  • Members
  • PipPipPip
  • 101 posts

Posted 31 March 2013 - 07:05 PM

Well the model may be overparaterized. but TMDD models are semi-mechanistic and do require more parameters to describe the PK. There is probably a good argument for or against but I'd prefer to at least test this hypothesis that its overparameterized.

 

I may have identified an alternate approach to simplify the model a bit. But that's what research is all about. i've got to rule out that there isn't something better out there ;)

 

I'd refer anyone interested to publications by Mager and Jusko on TMDD models for therapeutic proteins.

 

Elliot



#30 serge guzy

serge guzy

    Advanced Member

  • Members
  • PipPipPip
  • 485 posts

Posted 31 March 2013 - 07:10 PM

No problem, it is your model. please try with the corrected XF initial estimates (no boundaries) and put 0 as initial estimate (correspond to 50% F). Put only 5 iterations as it takes so much time to see at least that you get good fitting results.

Put 3000 samples, 5 iterations and MAP check box, no standard error. If you get results, do it again with the new results accepted and 5 iterations more with se's. The se's may take hours if it ever finishes (wait for the new version if you are patient) and want se's with QRPEM)

Good luck;

best

Serge



#31 Elliot Offman

Elliot Offman

    Advanced Member

  • Members
  • PipPipPip
  • 101 posts

Posted 31 March 2013 - 07:15 PM

By the way, can you also explain XF vs.F?

how would one interpret XF and 1-XF as a fraction, if its not a proportion totaling =1?



#32 Elliot Offman

Elliot Offman

    Advanced Member

  • Members
  • PipPipPip
  • 101 posts

Posted 31 March 2013 - 07:28 PM

would mind sending back the project via email?

I'm having problems loading my PC thinks it a zip file.



#33 serge guzy

serge guzy

    Advanced Member

  • Members
  • PipPipPip
  • 485 posts

Posted 31 March 2013 - 07:29 PM

XF is a normal distribution (going from -infinity to + infinity.

F is the fraction going in one of the compartment (I think we define SC absorption with F) and F is defined as equal to exp(XF)/1+exp(XF) . This term goes from 0 to 1 as XF goes from -infinity to + infinity.

 

The second fraction is 1-F and therefore 1- exp(XF)/()1+exp(XF))

 

The total is equal to 1 and each term cannot go above 1 and less than 0.

The only reason I used that transformation explicitly was that you wanted to sue QRPEM and QRPEM does not understand ilogit transformation when defining a structural parameter.

best

Serge



#34 serge guzy

serge guzy

    Advanced Member

  • Members
  • PipPipPip
  • 485 posts

Posted 31 March 2013 - 07:33 PM

Here we go. Model A. Try it and let me know.

best

Serge [file name=TMDD_model_with_parallel_absorption_official_version.phxproj size=1068745]http://www.pharsight.com/extranet/media/kunena/attachments/legacy/files/TMDD_model_with_parallel_absorption_official_version.phxproj[/file]

Attached Files



#35 Elliot Offman

Elliot Offman

    Advanced Member

  • Members
  • PipPipPip
  • 101 posts

Posted 31 March 2013 - 09:52 PM

For model A when I look at the initial estimate tab it doesn't seem to be inputting the 2nd dose point. I only see the Aa with the Tlag. I can shift that out to the right and see if moving but the zero order is not visible. Should we see some sort of constant input between time 0 to t


#36 serge guzy

serge guzy

    Advanced Member

  • Members
  • PipPipPip
  • 485 posts

Posted 31 March 2013 - 11:43 PM

Dear Elliott

 

For some reason, I see dosepoint2 instead of dosepoint .

 

It should be dosepoint, not dosepoint2.

best Regards;

Serge



#37 serge guzy

serge guzy

    Advanced Member

  • Members
  • PipPipPip
  • 485 posts

Posted 01 April 2013 - 12:05 AM

The model I see in Model A is not the model I sent you.

 

 

k0lymph=Alymph/dur is not correct as Alymph will change with time and you should write

k0lymph=dosing/dur where dosing is a covariate having the dose information.

I continue to check your code.

best

Serge



#38 Elliot Offman

Elliot Offman

    Advanced Member

  • Members
  • PipPipPip
  • 101 posts

Posted 01 April 2013 - 12:31 AM

for the covariate = dosing

does that mean I have 3 dosing colums?

(1) for Aa

(2) for Alymph

(3) for dosing?

 

All with the same dose since its a single fixed dose?



#39 serge guzy

serge guzy

    Advanced Member

  • Members
  • PipPipPip
  • 485 posts

Posted 01 April 2013 - 12:32 AM

Yes, it is in fact 3 copies of the same dosing information.

Best

Serge



#40 serge guzy

serge guzy

    Advanced Member

  • Members
  • PipPipPip
  • 485 posts

Posted 01 April 2013 - 12:42 AM

In fact you do not observe anything in the lymph, therefore putting just the zero order infusion into central is exactly the same mathematically and way easier.

You model is way too much overparametrized as you have only one dose and you cannot capture correctly the potential non linearity if there is one through this TMDD model (this is what I think).,

best

Serge






0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users