New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Visual predictive check interaval/prediction interal #104

Closed
moulirc opened this Issue Dec 6, 2017 · 7 comments

Comments

Projects
None yet
3 participants
@moulirc

moulirc commented Dec 6, 2017

output_forum

Hi,
I managed to do a simulation for meropena, when I did a visual predictive check the graph doesn't look good.. Please see attached picture and also have attached the project file.
Mero_run01.zip

@tobiasK2001

This comment has been minimized.

Show comment
Hide comment
@tobiasK2001

tobiasK2001 Dec 7, 2017

Member

Dear Moulirc,

it might look somewhat more conclusive if you are using log scale. And only simulate as long as you have data e.i. 4 h. Especially VPC and Confidence intervall plots could look less worse . But still the have gigantic unceartainty range of + - 4 log Scales. Iam not sure if this is a bug in the PI VPC tool... @msevestre could that be???
You might also think about not to use logP for fitting. In your PI LogP and TSspec are highly correlated...e.g. not identifiable. Indeed you get similar results only fitting TSspec, keeping logP as it is (e.g. use the one from drugbank) and maybe using another distribution model and charge dependend permeability as your compound is an acid as it seems.
But still even using only TSspec for fitting the VPC and confidance intervall looks horrible... (see attached file)
Maybe it could be good to use more data...
Hope this might help, Tobias
Mero_run01_edit tk.zip

Member

tobiasK2001 commented Dec 7, 2017

Dear Moulirc,

it might look somewhat more conclusive if you are using log scale. And only simulate as long as you have data e.i. 4 h. Especially VPC and Confidence intervall plots could look less worse . But still the have gigantic unceartainty range of + - 4 log Scales. Iam not sure if this is a bug in the PI VPC tool... @msevestre could that be???
You might also think about not to use logP for fitting. In your PI LogP and TSspec are highly correlated...e.g. not identifiable. Indeed you get similar results only fitting TSspec, keeping logP as it is (e.g. use the one from drugbank) and maybe using another distribution model and charge dependend permeability as your compound is an acid as it seems.
But still even using only TSspec for fitting the VPC and confidance intervall looks horrible... (see attached file)
Maybe it could be good to use more data...
Hope this might help, Tobias
Mero_run01_edit tk.zip

@msevestre

This comment has been minimized.

Show comment
Hide comment
@msevestre

msevestre Dec 8, 2017

Member

@moulirc Those are some weird looking graphs indeed.
@tobiasK2001 A bug is always possible... we are only humans after all :)

Member

msevestre commented Dec 8, 2017

@moulirc Those are some weird looking graphs indeed.
@tobiasK2001 A bug is always possible... we are only humans after all :)

@msevestre

This comment has been minimized.

Show comment
Hide comment
@msevestre

msevestre Dec 8, 2017

Member

@moulirc I agree with @tobiasK2001 analyses. Variation of Lipophilicity in such as huge range (-10 to 10) is not a good idea.
The correlation matrix shows that LogP and TSSpec are highly correlated thus rendering the results of your PI somewhat useless.
Moreover even when removing LogP from the PI, the error remains constant and the confidence interval are huge

image

All of which is a sign that this parameter has very little influence on your simulation.
In other words, those graphs are exactly displaying what they should. A very large uncertainty on those parameters.

Member

msevestre commented Dec 8, 2017

@moulirc I agree with @tobiasK2001 analyses. Variation of Lipophilicity in such as huge range (-10 to 10) is not a good idea.
The correlation matrix shows that LogP and TSSpec are highly correlated thus rendering the results of your PI somewhat useless.
Moreover even when removing LogP from the PI, the error remains constant and the confidence interval are huge

image

All of which is a sign that this parameter has very little influence on your simulation.
In other words, those graphs are exactly displaying what they should. A very large uncertainty on those parameters.

@msevestre msevestre added the answer label Dec 8, 2017

@moulirc

This comment has been minimized.

Show comment
Hide comment
@moulirc

moulirc Dec 11, 2017

Hi, @tobiasK2001 /@msevestre,

Please find attached the project file. Thanks for helping me out on this.
As I am bit naïve to PBPK still learning. I reduced the
Thanks heaps for your help.
mero

Mero_run2_corrected logp.zip

moulirc commented Dec 11, 2017

Hi, @tobiasK2001 /@msevestre,

Please find attached the project file. Thanks for helping me out on this.
As I am bit naïve to PBPK still learning. I reduced the
Thanks heaps for your help.
mero

Mero_run2_corrected logp.zip

@msevestre

This comment has been minimized.

Show comment
Hide comment
@msevestre

msevestre Dec 11, 2017

Member

What did you change? And what else do you need to know?

Member

msevestre commented Dec 11, 2017

What did you change? And what else do you need to know?

@moulirc

This comment has been minimized.

Show comment
Hide comment
@moulirc

moulirc Dec 11, 2017

oops sorry forgot to complete the sentence. I reduced the start value in PI and restarted the whole work as a new project and it worked.

kind regards
Mouli

moulirc commented Dec 11, 2017

oops sorry forgot to complete the sentence. I reduced the start value in PI and restarted the whole work as a new project and it worked.

kind regards
Mouli

@moulirc moulirc closed this Dec 11, 2017

@msevestre

This comment has been minimized.

Show comment
Hide comment
@msevestre
Member

msevestre commented Dec 11, 2017

@moulirc awesome

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment