NONMEM Users Network Archive

Hosted by Cognigen

Re: NONMEM code verification

From: Denney, William S. <William.S.Denney>
Date: Fri, 22 Jan 2016 14:03:35 +0000


Adding to Suruchi, one issue I have encountered in the past when working wi=
th complex or mathematically stiff models is that different integrators occ=
asionally give different results.

This will usually show up as instability of one of the integrators (big jum=
ps up and down on a percent basis from one point to the next). In the past=
 several years, this is getting rarer, but it can never be eliminated while=
 using computers and numerical methods.



On Jan 22, 2016, at 5:21, "Bakshi, S.D." <s.d.bakshi

Hi Ana,
What kind of weird behavior do you get?
If your model is complex in that it is nonlinear, it is likely that it show=
s multistationarity or oscillations or other emergent behavior which you ma=
y not expect. Some mathematical analysis of the structural model itself may=
 be useful in that case.


From: owner-nmusers
Sent: donderdag 21 januari 2016 22:42
To: Bauer, Robert
Cc: Nick Holford; nmusers
Subject: Re: [NMusers] NONMEM code verification

If working in R, you can use mrgsolve:<

It is open-source, validated R package interfacing with DLSODA solver in OD=

It should be able to directly utilize your NONMEM data set:<http=
(you might need to rename some columns ... use tolower() or mrgsolve has a =
function to do it).

If you're really making a really close comparison, pay attention to the sol=
ver tolerances and the number of digits in the answer (you can control them=
 in both NONMEM and mrgsolve). Caution: mrgsolve advances with the covaria=
te at T1 (in Bob's example). But I guess I'd tend to take that factor out =
of play and just look at predictions from some parameters constant over tim=
e where possible (assuming you're really wanting to diagnose what is happen=
ing with the ODEs; that's how I understood Nick's post).


On Thu, Jan 21, 2016 at 2:02 PM, Bauer, Robert <Robert.Bauer
The most common reason NONMEM may produce a different result than expected =
is if the evaluation depends on covariates listed in the data set that chan=
ges with each record, the default action of NONMEM is for an interval cover=
ing T1>Time<=T2, it uses the covariate at record TIME=T2. This behavio=
r can be changed with $BIND. Also, if you program in discontinuities that =
vary with model parameters, like changing a rate constant suddenly, you may=
 want to use the MTIME() system. Both of these items are discussed in viii=

Robert J. Bauer, Ph.D.
Vice President, Pharmacometrics R&D
ICON Early Phase
Office: (215) 616-6428<tel:%28215%29%20616-6428>
Mobile: (925) 286-0769<tel:%28925%29%20286-0769>

From: owner-nmusers
Behalf Of Nick Holford
Sent: Thursday, January 21, 2016 11:33 AM
To: nmusers
Subject: Re: [NMusers] NONMEM code verification


I call this process fixed effect (or deterministic) model qualification.

I code complex models using Berkeley Madonna then run NONMEM without any
$EST or $SIM records. The NONMEM PRED values should agree with the
Berkeley Madonna predictions if your structural model is coded the same
way in both systems.

Best wishes,


On 20-Jan-16 11:14, Ana Miranda Bastos wrote:
> Hi,
> I have a complex model with manually coded ODEs and multiple
> compartments. VPC etc seem ok but simulation results are getting a bit
> weird.
> I'd like to find out what people use to ensure that the NONMEM code
> written really represents the set of ODEs written on paper. Just to
> clarify, this is just to make sure the NONMEM instructions are
> actually a correct representation of the mathematical description of
> the model, not if the model is a correct representation of the biology
> at this stage.
> This problem is not so obvious when you use the built-in macros but
> once the model grows complex, and has a lot manual inputs, it is more
> likely that a bug creeps in.
> I'm looking for something more stringent than a code review by a peer.
> Thank you advance for your time and attention,
> Ana
> ------------------------------------------
> Ana Bastos, Pharm, MSc, PhD student
> Pharmacologie cellulaire et moléculaire
> (Cellular and Molecular Pharmacology Unit)
> Louvain Drug Research Institute
> Université catholique de Louvain (Catholic University of Louvain)
> UCL 7370 avenue E. Mounier 73
> 1200 Bruxelles, Belgique

Nick Holford, Professor Clinical Pharmacology
Dept Pharmacology & Clinical Pharmacology, Bldg 503 Room 302A
University of Auckland,85 Park Rd,Private Bag 92019,Auckland,New Zealand
office:+64(9)923-6730<tel:%2B64%289%29923-6730> mobile:NZ+64(21)46 23 53
email: n.holford<

"Declarative languages are a form of dementia -- they have no memory of eve=

Holford SD, Allegaert K, Anderson BJ, Kukanich B, Sousa AB, Steinman A, Pyp=
endop, B., Mehvar, R., Giorgi, M., Holford,N.H.G. Parent-metabolite pharmac=
okinetic models - tests of assumptions and predictions. Journal of Pharmaco=
logy & Clinical Toxicology. 2014;2(2):1023-34.
Holford N. Clinical pharmacology = disease progression + drug action. Br =
J Clin Pharmacol. 2015;79(1):18-27.

ICON plc made the following annotations.
This e-mail transmission may contain confidential or legally privileged inf=
ormation that is intended only for the individual or entity named in the e-=
mail address. If you are not the intended recipient, you are hereby notifie=
d that any disclosure, copying, distribution, or reliance upon the contents=
 of this e-mail is strictly prohibited. If you have received this e-mail tr=
ansmission in error, please reply to the sender, so that ICON plc can arran=
ge for proper delivery, and then please delete the message.

Thank You,

ICON plc
South County Business Park
Dublin 18
Registered number: 145835

Kyle Baron
Metrum Research Group
860-735-7043, Ext. 202

Received on Fri Jan 22 2016 - 09:03:35 EST

The NONMEM Users Network is maintained by ICON plc. Requests to subscribe to the network should be sent to:

Once subscribed, you may contribute to the discussion by emailing: