7 replies to this topic

[Elliot Offman]

is it logical to freeze only certain random effects and let other vary?

say in the instance you know that from prior information the inter-subject CV% on volume and clearance are ~15% and ~40% and you want to "freeze" the corresponding omega values, but you also want to estimate the random effects for ka, V2 and CL2 in a diagonal model.

 

(1) is this logical?

(2) how would this be coded in text form as I see there is only the option of freezing all random effects or none in the graphical interface or built-in models.

 

Thank you,

Elliot

[Samer Mouksassi]

Hi Elliot,

 

There is some applications that we want to fix one omega and estimate others.

Typically data will provide you enough information to estimate your clearance variability and volume and it is more difficult to estimate V2 and Cl2

 

Suppose you want to fix part of your omega you nee to break first your ranef block

 

By default it is one block and diagonal so you only have the option to freeze all.

 

In the GUI you can break into blocks/ rearrange the ordering and then for each block you can choose to freeze it. (Let me know if you will need help on this part)

 

Of course this can be also done in the text by using several ranef statements.

 

Samer

[Elliot Offman]

Yes, thank you, I do need some assistance here to ensure I'm coding correctly.

 

So here is a the ranef statement where all randome effects are estimated

ranef(diag(nV, nKa2, ntlag, nkint) = c(0.15, 0.58,0.033, 0.4))

}

 

what I really want is to freeze nV and nkint at values from prior research because my current data is based on n=4 animals, and is quite sparse, the estimates for these random effects are too low and much lower than reality would suggest. Therefore I want to fit the model and not estimate these 2 specific paramters.

 

 

I started with the built-in interface and split the blocks up. This is what I think I need to write, but would like confirmation.

 

ranef(diag(nV)(freeze) = c(0.15), diag(nkint)(freeze) = c(0.4), diag(nKa, nTlag) = c(0.58, 0.03))

}

 

Elliot

[Samer Mouksassi]

Dear Elliot,

yes the best way to learn PML is to use the GUI as a starting point.

Your code is good and should work

Bests,

Samer

[Elliot Offman]

Hi,

The model initiates however stops with the following error:

 

Index was out of range. Must be non-negative and less than the size of the collection.

Parameter name: index

[Samer Mouksassi]

Hi again,

Are you able to run your model using the graphical interface or the gui ?

if not you might have hit a bug affecting freezing omega in textual model and you might not run your model with the current released version.

Samer

[Elliot Offman]

This puts me in an awkward situation as at this point I only have access to the current version.

my model cannot be run in the graphical interface mode.

 

(1) can someone at Certara confirm indeed this is a bug?

(2) if you have access to the next version can I work offline with someone to test it with the un-released version?

(3) are there any other workarounds one of your colleagues at Certara may be able to assist with?

Elliot

[Linda Hughes]

Hi Elliott,

 

I can confirm that this was a defect in PHX 1.3 and that it has been fixed for the next release. The defect was publicized in a product notification (PN PHX014), and the product notification suggests one workaround for the PHX 1.3 release if the model must be kept as a textual model:

 

"Incorrect operation when text model is executed in edit mode and Omegas are frozen (QC 12067): When a text model is run with a frozen omega matrix where the freeze was entered in text mode, it completes with the correct answer (happens with all engines), but then fails upon returning to the UI with an 'index out of range' error. On the other hand, this does not occur if the model is a library (built in) model and the omega freeze is implemented by check box. The only known workaround if the user wants to execute the text model with that option is to execute the model in the Command-Line mode."

 

Let me know if this works for you or not, so that we know whether to pursue other options.

 

Regards,

Linda Hughes