Next Level Solutions is pleased to make available a suite of applications for Windows XP implementing parallel processing for data analysis in NONMEM.

Bootstrap
The first application available is Parallel Bootstrap for NONMEM. This application is designed for Windows SMP (symetrical multi processor) and multi core processors. This application is limited to Windows (2000 and XP). Parallel Bootstrap (Beta) requires a standard installation of NONMEM (with NMTRAN, NMLINK and NMFE5.bat), using the Intel, Compaq or G77 Fortran compiler. To install, download the zip file and extract the three files (setup.ext, setup.lst and bootstrap.cab) to any directory. Then run the setup.exe file.

This is still Beta, any any feedback/comments etc are appreciated (mark@nextlevelsolns.com).

Download Bootstrap Beta

NONMEM graphics
This is an Excel macro for making plots from NONMEM output ($TABLE output). It isn't a setup.exe, just a zip file. Unzip to c:\Program Files\NMMacros (navigate to the C drive, then use the "Use folder names" option in Winzip). There is an Excel file (nonmem_graphics.xls), a help file (NMGRAPH.hlp) and a readme.txt file
Download NONMEM graphics macro

Below is an Excel spreadsheet/macro that examines the effect of the sequence of subjects on whether NONMEM (version 6.0, with Intel 9.0 compiler) converges. This was presented at the 2006 ECPAG meeting. Anyone who uses NONMEM is familiar with the termination message "Minimization terminated due to rounding errors". But, what are the source(s) of these rounding errors, and what are they sensitive to? Next Level Solutions has investigated some issues related to instability and convergence in NONMEM. One possibility is in the summing of the partial likelihoods (contributions of each individual to the total likelyhood). The rounding error can depend on the order of addition for these numbers (theory and practical/background) . The practical background reference, I think can be very useful for those interested in general issues in numerical computation. The spreadsheet presents the results of 100 random sequences of subjects, with the OBJ, the values for THETA, OMEGA and SIGMA, and whether the model converged successfully (46%), and whether the covariance step was successful (covariance was not requested - it caused a crash that confused nmsee and the exce macro). Essentially (please see macros included in workbook for details), the original data was loaded into an array, by subject, then written out to a data file with the subjects in random order, then NONMEM executed. NONMEM sums the likelihoods in the order they appear in the data set, and so the order of summation in NONMEM was also randomly selected. Interestingly, while the parameters where not sensitive to sequence, whether the minimization was successful (vs terminated due to rounding errors) is sensitive to sequence. This is a control file/data set known to instability, and in fact was used as a test case for improving stability of NONMEM VI vs NONMEM V. Four significant digits were requested. The models/sequences that failed to minimize successfully generally had 3.4 to 3.8 significant digits (i.e., almost enough to minimize successfully). By sorting the contributions of each subject to the likelihood (sorting by absolute value) prior to summing, the results become independent of sequence of subjects, as well as resulting in more significant digits (4.3) and minimizing successfully. Work is ongoing to determine whether this improved stability makes a succesful covariance step more likely as well. Interestingly, the mean number of function evaluations decreased, and the time required for completion declined. (consistent with the Next Level policy of faster..better)
Excel spreadsheet/macro examining effect of sequence of subjects on convergence

NONMEM data set creation macro.

This macro can be used to create NONMEM data sets from simpler Excel spreadsheet.
Excel spreadsheet/macro for creation of NONMEM data sets


NMSEE2
This is an application/setup to display a summary of the output along side the actual NONMEM output. This is much like the application NMSEE , except it is a GUI. The real reason I wrote it is to capture output for the Genetic Algorithm application. I've tested it with about 30 different output file (different combinations of minimization success, covariance success, fixed and estimated THETA/OMEGA/SIGMA, block OMEGA/SIGMA, crashes etc. I'm hoping to get some feedback on more output files, to confirm that it appropriately captures all outputs. The application can be used as a free standing application, or put into the nmfe6.bat file (put the line c:\progra~1\nmsee2\nmsee2.exe %2 after the line COPY OUTPUT %2, near the end of the file. This will launch the application, displaying the output file and the summary.
For reference the criteria for the correlation test is that all offdiagonal elements of the correlation matrix have a absolute value less than 0.95, and the criteria for passing the eigenvalue test is that the ratio of the largest to the smallest eigenvalue is < 1000.
Download NMSee2.zip


DataCheck.xls
This is an Excel macro for checking NONMEM data sets. As with all of the stuff on this page, my motives are entirely selfish. I'm hoping to get some feedback on what other diagnositics people use for data checking, as well as feedback on this macro. The FDATA file is checked, along with the FCON file. To use it, first open DataCheck.xls in Excel. When the macro is launched (Tools..Macro..Macros..Datacheck.xls!DataCheck) you will be asked to select the FDATA file. It is required that the FCON file be in the same directory. You will then get a dialog with the format of the data and the first few lines of the data set (below). Currently, the macro is limited to 12 covariates (can be expanded if anyone needs it). Click "Run" and the following plots will be made: Histogram of each covariate Histogram of Dose times (time as in FDATA, with ADDL doses listed explicitly) Histogram of Dose Amounts Histogram of DVs Histogram of interval between doses (included the ADDL doses) On the worksheets (of each covariate, DVs, Dose times, Dose Amts and interdose intervals), the data outside the 95% CI will be highlighted yellow, those outside the 99% CI will be highlighted red. As usual: No warranty is implied - but please give feedback Very little error checking is included - but again please give feedback
Download Data Check Macro



Protocol for Evaluation of GA
Protocol for comparison of Automated search for model selection with traditional method
This protocol is currently under development for an objective retrospective parallel comparison of automated model selection algorithm vs traditional model selection algorithm. Comments are appreciated.
Download Protocol.doc


PPC
This is an application/setup to run Posterior Predicitive Check from NONMEM output. This is still in development, not even beta yet. Still, any input is welcome.
Download ppc.zip


Home