Home | Installation |
Control Streams | Bootstrap
| Randomization Test | Visual
Predictive Check | Autocovariate | Files | References
The Commands page provides a brief description of all key
WFN commands.
Nick Holford,
n.holford@auckland.ac.nz,
Department of Pharmacology & Clinical Pharmacology,
Last updated 21 June 2024 [Note: The latest changes to WFN are recorded in wfndown.htm]
WFN is a set of DOS batch command files and awk
scripts to help NONMEM users be more productive. The first version of WFN
(Wheels for NONMEM) was written in 1989 at UC San Francisco. The unix csh and
awk scripts evolved and were eventually ported to DOS in order to support
NONMEM as part of the RIDO
(Right Dose First Time) project.
Please see license.txt and gpl.txt. WFN is distributed under the terms of the GNU General Public License.
It uses gawk.exe which is also
distributed under the same terms.
A key concept in WFN is the idea of a run. A
run is based on an NM-TRAN control stream. The control stream filename (without its extension) defines the runname. NONMEM is
invoked using the nmgo batch command file with the runname as
its essential argument e.g. using the theopd.ctl control stream file:
nmgo theopd
You can execute nmgo from any directory you wish
but the NMTRAN control stream file (e.g. theopd.ctl) must be in the
current directory.
Note: Do not attempt to use a path to point to a
control stream in another directory. Long file names are supported but cannot
contain embedded blank characters.
nmgo creates a run directory for working files and
the final run results, processes the control stream then invokes NM-TRAN,
compiles, links and executes NONMEM. The NONMEM output listing is then
processed to provide a run
summary of the results.
Typical use of NONMEM involves several runs and
comparison of the objective function value (obj). The nmobj command
provides a simple table of runnames and obj values sorted in ascending obj order
to make it easy to compare the overall goodness of fit of each run. The output
of nmobj is stored in a file nmobj.txt.
Additional details can be obtained using the nmlst
command which presents the summaries for all the runs in the same order as nmobj.
These details are stored in a file nmlst.txt. Similar information is
presented one line per run using the nmmbt command. Results are
stored in nmmbt.txt in tab delimited format.
The results of the theopd run are presented in
summary form. The first 3 lines are column headings that display the parameter
names aligned with the estimates underneath. The 4th line contains
the run name (and nmout extension)
followed by the objective function value, estimation method, number of function
evaluations, significant digits, number of subjects, and number of
observations. Finally the NONMEM version number and PREDPP version number (if
used) are displayed.
Note that the ETASD and EPSSD are used to indicate
that the square root of the OMEGA and SIGMA values is displayed. In this
example a proportional model has been used for the variability of each
parameter so that an ETASD for E0 of 0.099 can be interpreted as an approximate CV of 9.9%. Empirical Bayes
estimate diagnostics test if the ETA average is 0 (ETAPval) and the percent
shrinkage of ETA (ETAshr%, EBVshr%) and EPS (EPSshr%). Shrinkage is shown using
the standard deviation method. The relative standard error of the random effect
estimates (OMEGA and SIGMA) are expressed relative to ETASD and EPSD. The
format can be modified using NMRAW.
THETA: POP_E0 POP_EMAX POP_EC50
ETA: PPV_E0 PPV_EMAX PPV_EC50
ERR: RUV_SD
theopd.lst 5801.321
FOCE eval=295 sig=+4.2 sub=153 obs=574 NM7.4.0
THETA = 139 191 8.8
ETASD = 0.099 0.268 1.463
ETAPval = 0.346
0.000 0.605
ETAshrSD% = 81.0 55.8 35.6
EBVshrSD% = 81.8 59.6 29.9
EPSshrSD% = 9.3
EPSSD = 81.055
THETA:se% = 6.9 9.4 14.9
ETASD:se% =
168.6 20.7 16.9
EPSSD:se% = 5.3
MINIMIZATION
SUCCESSFUL
Ttot 0:3.87 Test 0:1.7 Tcov 0:0.73 Ttcl 0:1.42
The last line of the run summary shows the total
execution time (Ttot), NONMEM estimation time (Test), covariance time (Tcov)
and the time for the control stream translation, compile and link part of the
run (Tcl). Times are formatted as minutes:seconds for on-screen display and
minutes and fractions of minutes in the run.smy
file and the output of nmmbt. Timing
and results will depend on the computer and compiler (results above are with 32
bit gfortran).
The run directory is used as a working directory for
NM-TRAN and NONMEM temporary files. These files are deleted at the end of a run
unless the Windows Commmand environment variable nmclean is set to
"n". The run directory name is made up from the runname and an
extension which by default is .nm7 e.g. theopd.nm7.
Note: If the WFN compilation size system is used then
the run directory name will reflect the compilation size e.g. if the regular
size is used the run directory will be theopd.reg.
The run results files saved at the end of each run
are:
Runname.lst |
NONMEM output listing (see nmout) |
Runname.smy |
Run summary (one line per problem or sub-problem) |
Runname.smr |
Run summary with control stream appended |
Runname.log |
A log file which may contain information about errors if a problem arises |
Runname.ctl |
The control stream used by NM-TRAN (see nmctl) |
Runname.fit |
Table file if $TABLE NOPRINT is specified (see nmtbl) |
Runname.msf |
MSFO file if MSFO is specified |
nmpathlist.txt |
Paths
used by compiler and MPI |
.coi,.cor..cov,.cpu,.ext,.phi,.shk,.shm,.xml |
Miscellaneous
NONMEM output files |
Runname_mpi.log |
message
passing interface log file if mpi is used |
mlxgo and mlxbs commands can be
used to run Monolix (tested with version 4.2.2) if a Monolix control file is
provided instead of an NM-TRAN control file. Control files with the extension
.mlxtran will be passed to Monolix. Monolix control files may refer to data,
model, algorithm, and graphics files. In addition to standard Monolix output a
WFN smy file is created. The mlxmbt
command may be used to summarize Monolix results.
The mlxbs
and nmgosim commands gives Monolix
users the opportunity to perform non-parametric and parametric bootstraps. For
parametric bootstraps the simulated data is created using a NM-TRAN control
file and the estimation step is performed using a Monolix control file (with
.,mlxtran extension).
wfn.bat has to be modified to set the MONOLIX and MONOLIXDATA paths. The
username in MONOLIXDATA should be a Windows user name. See %WFNHOME%\run\wfn.txt for an example of where to set these
variables.
set MONOLIX=C:\Monolix\Monolix422\Monolix\monolix422s
set MONOLIXDATA="C:\Documents and
Settings\username\monolixData"
Monolix examples are provided in the %WFNHOME%\run\mlx
folder. The batch file gomlx.bat may
be used to run NONMEM and Monolix examples using the same data and similar
models.
Note that data files for use with Monolix by default
are assumed to be in comma separated value (CSV) format. A different separator
can be defined using the NMSEP environment variable e.g.
set NMSEP=;
will use the semi-colon as a separator in the data
file.
A list of WFN commands and how to use them can be
found in Commands.
Home | Installation |
Control Streams | Bootstrap
| Randomization Test | Visual
Predictive Check | Autocovariate | Files | References