MonolixSuite 2021R1 Release Notes
December 2021
The document is the release notes for MonolixSuite2021R1 and contains most of the evolution and bug fixes of the software along with the data set and Mlxtran management.
Global display
Enhancements:
* The global display of all softwares has been improved.
* In all applications, renaming and ordering of subplots after a split is now possible.
* In all applications, user-defined ticks on plot axes can be defined.
* In PKanalix and Monolix, a covariate summary is now displayed in the data tab.
* In PKanalix and Monolix, in the observed data plot, there are additional options for the trend line (arithmetic or geometric mean) and error bars (standard deviation or standard error).
* In Monolix and Simulx, the model editor is now embedded in the application.
* In all applications, the display of the models of the libraries is much faster.
* Pop-up warning windows asking for confirmation to run or make some edits, with a button “don’t show again”, which can also be restored in the preferences.
Bug fix:
* The option “–disable-gpu” was not working anymore in the 2020 version.
Data set
Enhancements:
* It is now possible to set several columns of type “ignored lines”.
* In PKanalix and Monolix, a fingerprint of the dataset is saved at the time of the run. If the content of the data set file changes, the results will not reload.
==Bug fix:
* The data interpretation was incorrect in the case of occasions that are not grouped together (i.e lines with occ=1 are not next to each other in the data set). The bug appears when there is a dose line and an observation line with the same time, followed by another dose at a later time, and if for the dose and observation at the same time, the dose comes first.
* In case of occasions, the Monolix GUI was freezing when setting a column as “ignored observation” was ignoring all data points of one subject-occasion.
* Value ‘NA’ in the amount column is now handled as ‘.’. Previously it was ignoring the value of the Observation column on the same line.
* The complement of a filtered dataset was actually the complement of the parent dataset.
Mlxtran
Enhancements:
* There is a new parent-metabolite library. It uses analytical solutions when the model is linear.
* Linear models with sequential occasions (not separated by a washout) are now computed using the analytical solution and are thus much faster. Note that linear models with washout-separated occasions were already using analytical solutions in the previous version.
* Additional models in the PK, PKPD and TMDD libraries: 1) lag time in case of bolus or infusion, 2) zero-order absorption in case of bolus/infusion + extravascular.
* It is now possible to combine a first-order absorption model with an infusion in the data set.
Bug fix:
* When there were two successive changes of sign in the model (e.g ddt_E = -(-kin+kp) – kg*E), the compilation failed.
* There was an error when loading a model file starting with <MODEL>.
PKanalix
Enhancements:
* There is a new bioequivalence module.
* The split of the summary tables for the CA and NCA parameters is now more flexible. It is possible to split and filter by continuous and categorical covariates as well as the acceptance criteria. Covariate groups can be customized and are displayed as separate columns in the table.
* A new output file with the points included for the lambda-z calculation has been added.
* It is possible to define default units in the preferences.
* There is a new plot corresponding to the lambda-z regression for the NCA. This plot can also be used to display the observed data with one id per subplot.
* It is now possible to overlay the points on top of the boxplots to have a better feel of the distribution.
* For IV bolus, points at Tmax = 0 are now considered for inclusion in the calculation of the lambda-z. They were previously always excluded.
* The determination of the points included in the lambda-z has been modified: once the set of points leading to the maximum adjusted R2 has been found, adjusted R2 values for sets with larger numbers of points are now compared only to the maximum adjusted R2 value.
* A project with infusion + steady-state has been added to the demos and to the validationSuite.
Bug fix:
* In the NCA summary table, NTOT and NMISS were not properly calculated after a split or filter.
* In the IV bolus case, when there is no observation at the time of the dose, the two first observations are used to back-extrapolate C0. When one of the two observations was negative, PKanalix was returning an error instead of taking the first observation as C0.
* The units of the NCA parameters divided by the dose (e.g AUCINF_D_obs) were wrong.
* Vss_obs and Vss_pred were not displayed in the result table in case of single-dose (non-steady-state) data.
* In case of several observation ids, the observation names (e.g “y1”) were not well updated when changing the “obs id to use” in the data tab.
Monolix
Enhancements:
* [Algos] The possibility to fix the value of the correlation parameters has been added.
* [Algos] More stable estimation of the correlation parameters (correlation parameters do not tend to go to 1 or -1 anymore).
* [Algos] It is now possible to put a bayesian prior on the omega parameters.
* [Algos] Use of the Nelder-Mead algorithm (for instance for parameters without variability) can now be distributed (parallelization over several machines with monolix MPI).
* [Initial estimate tab] The auto-init button is now available for any model, even custom.
* [Initial estimate tab] Parameters which are fixed are colored in orange to indicate that they won’t be modified by the auto-init procedure.
* [Settings] It is now possible to set all SAEM iterations to zero in one click in the SAEM settings.
* [Results] Summary information about the task iterations in the results is now displayed, saved and reloaded.
* [Proposal] We now propose the correlation at IOV level in the proposal.
* [Plots / VPC] VPC for TTE is now created even if some simulations lead to NaNs. The percentage of NaNs is displayed.
* [Plots / VPC] The export of the VPC simulations is now available as a preference in addition to the export button, to allow exporting the VPC simulations automatically after each run, and also when running from the command line.
* [Plots / VPC] There is a new option to plot the x-axis as “time after the last dose”.
* [Convergence assessment] The settings for the convergence assessment are now saved and reloaded.
* [Model building] The settings for the model building are now saved and reloaded.
* [Model building] A button to reset the settings to their default value has been added.
* [Model building] The covSamba-COSSAC method (which was running one iteration of samba followed by the usual cossac algorithm) has been replaced by the covSamba algorithm (called SAMBA in automatic covariate model building tab), which applies the proposal for the covariate model several times until no further change is suggested. The same options regarding the parameters and covariates to test as for cossac and scm are available.
* [Run] When running the project will overwrite existing results, a warning pop-up window appears and asks for confirmation.
* [Model building] The default p-values threshold for the forward LRT for COSSAC has been changed to 0.01. Having different p-values in forward and backward for COSSAC makes little sense.
* [Convergence assessment] The runs of the convergence assessment can now be reloaded as usual runs.
* [Convergence assessment and model building] The folder structure in the result folder has been modified to be clearer.
Bug fix:
* [Data] Setting a covariate column as “IGNORE” in the data tab was removing unrelated transformed categorical covariates in the “Statistical model & Tasks” tab.
* [Initial estimates] Negative betas were not well updated in the GUI when doing “use last estimates” or when typing a value. Note that in the 2020 version, the value was nevertheless properly taken into account for the calculations and saved.
* [Plots / Preferences] The colors were not updated when switching to a scheme with fewer colors.
* [Plots / Preferences] The group highlighting was not working with multi-covariate groups.
* [Plots / Stratify] When stratifying on grouped discrete covariate, the application was crashing.
* [Plots / Indiv fits] Log-scale sometimes applied to the current page only instead of all pages.
* [Plots / Observed data] If individuals were selected via individual selection in Stratify, the mean and error bars were not properly displayed.
* [Plots / Scatter plot of the residuals] Censored data points had wrong times displayed in the tooltip when hovering over the point.
* [Plots / BLQ predictive check] On the y-axis, the fraction of observations being BLQ was wrong.
* [Plots / KM plot] If all individuals within a subplot were censored, the censored events were not displayed.
* [Plots / VPC] In case of count or categorical data VPC, the yLabel was “P(undefined=0)”, using “undefined” instead of the model variable name.
* [Plots / Prediction distribution] In case of occasions not separated by a washout, a crash has been fixed.
* [Plots / All] In the stratify panel, the filtering choices are now reset when the corresponding covariate categories are modified.
* [Results] When SE and RSE could not be estimated, the population parameter estimates table was showing an empty cell instead of NaN. Note that the NaN were properly shown after reloading the project.
* [Results] RSE for covariates effects with negative estimated value for the covariate effect were displayed with a wrong negative sign in populationParameters.txt. Note that the value in the GUI and in the summary.txt file were correct.
* [Results] Values “1” on the diagonal of the correlation matrix of the estimates were sometimes incorrectly colored in red.
* [Results] The likelihood calculations via importance sampling when individuals where present in the data set but having no observations (e.g in case of PK/PD data set with placebo individuals with PD data but no PK to which a PK model only is fitted) were wrong.
* [Results] The VPC simulation export in case of censored data was not properly handled for the censored data. The simulated value was indicated as censored if the value in the original data set was below the LOQ, instead of the simulated value.
* [Results] Small differences in results appearing in rare situations when running via the GUI or via the lixoftConnectors have been fixed.
* [Results] The conditional mean values in the individual parameter table were calculated by doing the mean in the non-gaussian space (e.g mean(V_i)). This was inconsistent with the value used in the plots and for the predictions in the predictions.txt file. The mean is now done in the gaussian space (e.g exp(mean(log(V_i))) for a lognormal parameter).
* [Convergence assessment] The convergence assessment does not stop anymore if one of the runs generates an error.
Sycomore
Enhancements:
* The load of the Monolix runs into Sycomore is now much faster.
* When the project has not run yet, the initial parameter values are displayed in the comparison table. They are now in gray to indicate that they are initial values and not final estimates.
Bug fix:
* In the comparison table, labels corresponding to the non-default task settings were unclear.
Simulx
Enhancements:
* [Import] When importing a Monolix project, a new population parameter element is automatically created. It allows to sample population parameters from their uncertainty distribution defined by the variance-covariance matrix of the estimates (calculated in Monolix via linearization or stochastic approximation), when simulating several replicates.
* [Import] When importing a Monolix project, the created element mlx_CovDist now uses lognormal distributions if all covariate values of the Monolix data set are positive.
* [Export] It is now possible to export the simulated data as a MonolixSuite formatted data set, which can easily be loaded in Monolix or PKanalix.
* [Exploration] It is now possible to overlay the data of an external MonolixSuite formatted dataset on top of the simulations in the Exploration tab.
* [Exploration] In the Exploration tab, the user can now set the y linear- or log-scale independently for each subplot.
* [Exploration] Custom axis limits have been added for each subplot.
* [Exploration] “Pitch wheel” sliders are now available to modify the values of the individual parameters.
* [Outcomes] Additional options to define outcomes have been added: the first and the last value of the output.
* [Run] The memory usage has been decreased.
* [Run] When running replicates, a progression bar indicates the simulation progression.
* [Plots] In case of replicates, in the Individual output plots, the selection of the replicates to show is now possible for all types of data (continuous, count/categorical, and TTE).
* [Definition] For treatment elements with non-compliance option, the non-compliance information is now displayed when clicking on “view”.
Bug fix:
* For categorical output with categories not starting at 0, the simulated values were shifted and starting at zero.
* The Wilcoxon rank test p-value calculations for TTE outcomes were wrong.
* After importing a Monolix project with count, categorical or TTE observations, output elements corresponding to these observations were not selected by default in simulation. These outputs were thus missing when running simulx(project=…) with RsSimulx in R.
* The exported file regressors.txt was incomplete. This also impacted the regressor outputs when running simulx() with RsSimulx.
* Typing negative values with many digits was truncated in both display, save and values used for simulation.
* Covariate categories visible in Stratify were not properly updated when the number of categories used in the simulation was changed and the project was rerun.
* The occasion header was missing in some output files, leading to a mismatch between the column content and the header.
* In case of occasions, and occasion-wise definition of covariates, changing a categorical covariate category choice was sometimes not properly taken into account.
* In the model, random variables defined with “distribution=geometric” were returning NaN.
* A crash appearing in the case of output elements differing between the simulation groups has been fixed.
* After importing a Monolix project, if an “additional line” was added to the model before doing a “save/save as”, upon “save” the path to the model file was not correct (still pointing to “temp” instead of the result folder).
* In the Exploration tab, when using an indiv param element with several individuals, if the ids were strings (not integers), they were not properly displayed in the individual selection slider.
* After reloading a Simulx project which was created using an “import from Monolix”, doing a “save as” was not properly copying the model file to the new result folder.
Lixoft Connectors
Enhancements:
* There is a function to get the plots as ggplot object for all Monolix and PKanalix plots.
* There are new functions getConsoleMode/setConsoleMode to set the level of console output. This allows to follow the progression of the run in the console.
* The stratification covariates and transformed covariate values were added in the getCovariateInformation() results.
* Units information were added to the NCA results.
* When the information to be returned by getSimulationResults() is too large, a warning message indicates it.
Bug fix:
* The project creation with newProject() in case of discrete data and mapping between the model outputs and the observation ids was returning an error.
MlxEditor
Enhancements:
* The MlxEditor application has been completely re-implemented.
* A clearer indication of the lines in case the syntax check detects an error.
* Auto-completion is available.
* It is possible to save several files into a mlxEditor session.
