Release Notes MonolixSuite 2024R1
March 2024
These release notes outline the key enhancements, new features, bug fixes, and usability improvements introduced in MonolixSuite 2024R1. Users are encouraged to explore the updated functionalities and provide feedback for further enhancements.
NEW FEATURES:
Installation and Licenses
The default installation folder on Windows has been changed to C:/Program Files/.
The requirement of an X server for lixoftConnectors usage has been eliminated.
New installer with reduced size, integrated silent installer, and possibility to activate the license in the command line.
General Project Management
A new pop-up window under Help > About License now displays license information within the interface.
The addition of a “Share Project” button facilitates the creation of a zip file containing all necessary files for project reload.
Warning pop-up boxes have been implemented to alert users when launching a run with unsaved projects to prevent inconsistent results, or when saving if previous results may become inconsistent.
General Display
Qt was upgraded to Qt6, allowing users to zoom in the GUI using the mouse wheel while holding the Ctrl key.
Plots – Global
Improvement of plots customization with the inclusion of plot presets for plot settings, stratification and preferences, allowing to:
easily transfer plot customizations to other plots within the same project,
save plot customizations as custom presets and apply them across projects,
apply a built-in publication-ready preset for achieving standardized report-like plots with a single click.
Hex codes can now be used to choose colors in plots.
There is a new zoom setting for the plots, that automatically changes the size of all elements.
New information message in the Plots tab when there are valid results and the Plots task has not been run.
Plots – Monolix & PKanalix
[Observed data] Toggle added to hide/display censored data and use them in statistical calculations.
[Observed data / VPC] Options to display time since last dose, nominal time, or a regressor on the x-axis.
In Monolix, the ability to save multiple versions of each plot in the UI (e.g. in log and linear scale).
In Monolix, plots data can now be saved as binary files in the results folder, and previously generated plots can be automatically reloaded when loading a project. This behavior can be enabled as a preference.
Data Set – Global
Significant reduction in project reload time for large datasets (on average 2 to 3 times faster) by introducing a new option to save datasets as binary files.
Expanded compatibility to load xls, xlsx, sas2bdat, and xpt files in addition to csv and txt.
In the Data Formatting module, new option to apply treatments to a group of ids.
In the Data Formatting module, when “dose as occasions” is checked, new option “Duplicate observations at dose time into each occasion”.
Possibility to save and manage Data Formatting presets that can be easily applied to a new project.
New “Format” button in the Data tab to send the dataset to Data Formatting.
Introduction of the NOMINAL TIME column type, that can be used on the plots of Observed data as well as in PKanalix NCA data plot and table of concentrations.
Continuous covariates having the same value for all individuals are now kept in the GUI and can be used in the model (we recommend fixing their covariate effects as they are not identifiable).
For IGNORED LINE column-type, all integers values are allowed instead of only 0 and 1.
Consistent data tagging is kept after any change to the dataset.
New shortcut to send current header types tagging to preferences.
Added default column headers in Preferences for ID (subjid), OBSERVATION (aval), INTERDOSE INTERVAL (trtrint), TIME (afrlt , arrlt) and NOMINAL TIME (nfrlt, nrrlt, ntime, nominal_time, nomtm_d, nomtm_h).
Model selection and prediction – Global
Piecewise analytical solutions are used with time-varying regressors.
Option for linear interpolation of regressors.
Introduction of new settings odeAbsTol and odeRelTol that can be used in the structural model to modify the ODE solver tolerance.
Library filters are kept when changing the structural model selection from a model from the PK library to a model from the PKPD or Parent Metabolite library.
In Monolix and PKanalix, notification above the structural model when it is defined with macros and an analytical solution will be used.
PKanalix
Improved user interface for selecting NCA parameters to compute.
New handling of sparse data for NCA with:
automated calculation and visualization of mean concentration profiles,
option to stratify profiles based on categorical covariates,
extended settings to handle censored concentrations.
Ability to compute individual ratios of NCA parameters between occasions, e.g. AUC(test)/AUC(ref).
Ability to compute custom NCA parameters as formulas of existing parameters and covariates, and save them in Preferences.
New concentration-time table with options to:
display individual profiles or summary statistics,
stratify the table across covariates such as days, periods, cycles, and treatments,
use actual time or nominal time.
Indication of compliance to general rule for lambda_z estimation in the individual results.
Addition of a new “NCA data” plot displaying concentrations used for NCA.
Option to calculate steady-state NCA parameters after each dose even without SS=1 column.
Addition of new units: IU, mIU, uIU, GBq, MBq, kBq, Bq, dL, mgEq, ugEq, ngEq, pgEq, fg.
Allow datasets and profiles (i.e. subject-occasions) without dose amount.
Improved CA autoinit for models from the PK library, with a pre-initialization of parameters based on NCA-like analytical expressions.
Monolix
Inclusion of non-parametric and parametric bootstrap functionalities in the user interface and via the lixoftConnectors, with:
option for stratified resampling based on categorical covariates,
custom confidence interval level, sample size, initial estimates,
reporting of the bootstrap confidence intervals next to the estimation results,
real-time display of bootstrap estimates and their statistics,
real-time plot of median and confidence intervals over the bootstrap estimates,
distribution plots of bootstrap estimates and OFV,
option to save and load bootstrap runs,
possibility to stop and resume bootstrap or add more bootstrap runs,
possibility to replace or filter bootstrap runs with failed convergence,
command line distributed calculation on a cluster.
Shrinkage calculation is now based on standard deviation instead of variance.
Shrinkage is now displayed in summary tables of individual parameters and on the plot of distributions of the random effects, and it is included in the table of population estimates exported in the default report.
Option to display typical values of population parameters for different categorical covariate modalities in the user interface.
Calculation and display of 95% confidence intervals on population parameter estimates.
Ability to click “use last estimates” for parameters that were previously estimated even after changing the model or data.
Enhanced conversion of RSEs from Gaussian to non-Gaussian domains, using an exact formula for log-normal parameters and numerical integration for logit and probit, instead of Jacobian approximation that is approximate for highly non-linear transformations and high RSEs.
Decreased tolerance of the ODE solver used in the calculation of the Fisher Information Matrix for improved standard error calculations when using the non-stiff ODE solver.
Optimization of MCMC settings for faster population parameter estimation (7% on average).
New warning when the user calculates the standard errors or likelihood by linearization when the EBEs have not been run.
Sycomore
Added functionality to export the tree diagram as an SVG image.
Export of Monolix projects to Sycomore.
Simulx
The 95% confidence interval on the power of the study (percentage of successes over replicates) is now reported.
Option to import unused covariates from Monolix or PKanalix for simulations (e.g. to define a weight-based treatment even if weight is not a covariate in the model).
Option to define regressors as distributions (e.g. when simulating from a PD model with individual PK parameters as regressors, the PK parameters can be sampled from a distribution).
Button to automatically add an [INDIVIDUAL] block in the model with log-normal distributions and inter-individual variability describing all parameters.
Improved reporting of NaNs in simulations with the percentage of simulated individuals with NaNs.
Addition of a toggle to display only median predictions in simulation output distribution plots.
Reporting
Inclusion of a new keyword for project file path.
Default reports generated with Monolix have an improved table style.
Lixoft Connectors / Command Line
Parallelized convergence assessment on a cluster from the command line.
Possibility to specify all settings of model building run in the command line with a config file.
Enhanced documentation for Monolix and PKanalix connectors to provide more guidance and working examples.
Addition of new functions printSimulationSetup(), displaying everything that is set for simulation, and printOutcomesEndpoints(), displaying all defined outcomes and endpoints.
New warning when versions of lixoftConnectors and MonolixSuite do not match.
importMonolixProject() has been removed from lixoftConnectors (deprecated function, importProject() should be used instead).
BUGS FIXED / USABILITY PROBLEMS:
Installation / Licenses
The interface of the validationSuite for PKanalix or Simulx could not run without a Monolix license.
There was a crash in R occurring when the pop-up window requesting a license to use the lixoftConnectors was closed without providing a license.
The lixoft folder inside the temp directory was not deleted after installation.
General Project Management
The results folders for model building and convergence assessment were not copied with “save as”.
General Display
Removed the “high dpi scaling” preference (deprecated in Qt 6 version).
Added “disable-gpu” by default in the .sh and .bat files to launch MonolixSuite apps to prevent GUI issues.
Fixed visibility issues for lists of models from some model libraries on small screens.
Enhanced visibility of the search field in demos.
Tables with invalid results in the GUI were disappearing when changing a table display setting.
Column header width did not adjust correctly when adding significant digits.
Fixed icon resolution on Mac.
In Monolix, the files populationParameters.txt, predictions.txt, and summary.txt were not removed from the results folder when overwriting a project with results with another project without results.
Plots – Global
Plot settings and preferences are now kept after running a task or after changing the data or model.
[Observed data] Preferences were not properly applied to error bar labels.
[Observed data] Improved accuracy in the calculation of statistics on number of observations and doses in Observed data plots: they are now calculated only on the subjects shown on the plot and have more significant digits.
[Observed data] The “Merged split” option was not properly untoggled when “mean” was unselected.
[Observed data] There were freezing and shutdown issues in PKanalix or Monolix when navigating through IDs in the selection panel of stratify if an ID had only censored data.
[Observed data / Individual output TTE] If at least one individual in a time-to-event dataset has two observations (even 0) at the same time, the survival estimate (Kaplan-Meier curve) was not adjusted to take into account the right-censoring events (dropouts).
[Observed data / Individual output TTE] Events occurring at the first time in the dataset (usually time 0) were missing from the Kaplan-Meier curve.
[Observed data] In PKanalix, the displayed total number of subjects in case of occasions was not well calculated and corresponded to the total number of subject-occasions.
[VPC / Output distribution TTE] With right-censored events, in some cases the Turnbull estimator was not well calculated during the last censoring interval: the stopping criterion in the algorithm is now set to 1e-6 instead of 1e-4 and the maximum number of iterations is 1000 instead of 200.
It was not possible to change the color of a stratification group combined from several covariates after project reload.
Setting a custom color for histogram bar strokes did not work.
Reporting – Global
If a project file path contained more than 168 characters, generating a report could fail on Windows.
Report generation was frozen if an invalid setting value was written in the placeholder.
If the axes in the bivariate data viewer had been modified (to select which observation goes on which axis), default report generation could freeze.
Data Set / Data Formatting
Improved the use of space and commas in censoring tags.
When sorting columns in Data Formatting, a different line than the header turned blue.
In Data Formatting, the header lines did not stay on top while sorting.
Improved the behavior of the warning message about ignored columns in external treatment files from Data Formatting.
Data Set / Data
Projects with double quotes in covariate modality names had reloading issues.
In specific cases, multiple additional covariate header types were incorrectly ordered.
There was an invalid error message when canceling data edition with a regressor column.
Changing column tagging from LIMIT to IGNORE and then canceling led to an error.
The warning message about missing covariate definitions had incorrect line numbering.
Improved visibility of the warning when no TIME column is defined.
The case “two different regressor values at the same time for the same id” was not always properly detected. When importing a project with this case into Simulx, the regressor element was not created without any error message.
Datxplore
After exporting a Datxplore project with a filtered dataset into Monolix, the Filters and Data Formatting tabs were empty.
There was an error occurring when changing the observation column in an unsaved Datxplore project imported from Monolix.
There was a crash when importing a PKanalix or Monolix project with constant covariates.
PKanalix / Tasks
[Check lambda_z] Improved the handling of null concentrations in lambda_z regression: they are now ignored when the selected rule is to use a fixed number of points.
[Check lambda_z] There were display issues in “Check lambda_z” plots in case of “#” in id names.
[Check lambda_z] Improved performance of Check lambda_z plot with many individuals.
[NCA] Subjects with only BLQ samples were not handled following the BLQ method before Tmax chosen by the user.
[NCA] Resolved inaccuracies in AUC_TAU and partial AUC calculations in case of very small extrapolated concentrations.
[NCA] Partial AUCs with overlapping time intervals were incorrectly calculated.
[NCA] The “% of extrapolated AUC” acceptance criterion can now be used based on AUCINF_obs or AUCINF_pred, instead of only AUCINF_pred.
[NCA] When requesting two partial AUCs with very close times, some of them could be missing due to rounding of times in the parameters names. Partial AUC parameters names are now rounded with 12 significant digits and cannot exist several times.
[NCA] NCA parameters aliases can now be used also in R and the plots generated in R.
[NCA] Improved the performance when loading projects with many NCA parameters, such as partial AUCs.
[NCA] There was a crash when running NCA again in a project where the NCA summary table was split by a covariate and where some parameters were not displayed in the table for some modalities because of filtering, and saving the project led to an error.
[BE] Values in the bioequivalence table in the GUI are now rounded according to the “number of digits” preferences instead of always 6 digits.
[CA] Scaling factors were wrongly applied to CA predictions after exporting a project with a scaling factor in the units to Monolix.
PKanalix / Reporting
In case of multiple observation types, plots of Observed data were shown in the wrong order in the default report generated by PKanalix.
Monolix / Model & Tasks
[Tasks] Shrinkage was calculated using the standard deviation of (for example for a log-normally distributed parameter without covariate effect) “log(mean(psi_i))-log(psi_pop)”, we now use the standard deviation of eta_i_mean instead.
[Structural Model] It is no longer possible to define in a model a compartment with the concentration keyword, but without volume.
[Structural Model] It was possible to set a model not compatible with the data, and an error was raised without preventing the action. This could lead to a crash if setting a single event TTE model with a repeated event dataset.
[Structural Model] In the 2023R1 TGI library, “Ribba” and “two population” models with saturations could not be loaded because of a syntax error.
[Check initial estimates] For parameters set as “fixed”, autoinit was also optimizing their value even if the value shown in the GUI did not change.
[Check initial estimates] There was a crash in Check initial estimates when a dataset had been accepted again with no change.
[Statistical Model] Modalities of categorical covariates were not properly updated in the statistical model if some of them were removed by filtering the dataset.
[Scenario] A warning was added when results from a task are deleted after running a linked task.
[Scenario] When the “Plots” task was disabled to avoid generating plots that are not consistent with project settings, it was incorrectly re-enabled after changing the “number of simulations” setting.
[Convergence Assessment] After running the convergence assessment on a Linux machine, the project did not reopen on a Windows machine and the error “is not a valid assessment type” appeared. Convergence assessment runs done on Linux now load correctly on Windows.
[Model Building] In Model Building, variables without random effects were not re-enabled for COSSAC and SCM after clicking SAMBA.
[Model Building] In Model Building, the “load” button next to each model now loads the corresponding project without having to save it again.
[Model Building] When using SCM and the likelihood ratio test (LRT), in case two added covariates were both leading to a p-value so low it was considered as zero, then the first covariate tested was selected instead of the one leading to the strongest LL improvement.
[Model Building] When the likelihood ratio test (LRT) was used, when considering categorical covariate with more than 2 categories, the degrees of freedom was always considered to be one, instead of considering the number of additional parameters (betas) to estimate.
The linear ode solver (used with odeType=linear) has been removed.
Monolix / Plots
[Observed data] In Monolix, the default minimum number of data per bin in the mean curve has been set to 5 instead of 10 to make the absorption phase easier to capture.
[Individual fits] There was a crash when generating an additional covariate based on observation numbers after the plots were generated, and using it to sort the plots of individual fits.
[Individual fits] Plots preferences for population fits with typical or individual covariates have been renamed to be clearer, as well as the theoretical median for the distribution of standardized random effects.
[Individual fits] There was an error when an individual with non-overlapping occasions had missing observations in one occasion.
[Observations vs Predictions] When several observation ids are present, the “90% prediction interval” and “outlier proportion” were calculated using the error model of one of the observation id for all others observation ids.
[Observations vs Predictions] When there is censored data, the simulated censored samples in the plot of observations vs population predictions are now always based on the conditional modes if they have been estimated, and the conditional mean only otherwise. In addition the column Y_simBlq_popPred has been removed from the charts data for this plot as it was never used.
[VPC charts data] The file with exported VPC simulations was removed when exporting charts data.
[VPC TTE] The default y-axis limits of the TTE VPC are now always [0,1].
[VPC] Prediction intervals did not correctly adapt when the axis limits were changed.
[VPC] Plots simulations were generated with different random number generator initialization after a run or after a load.
[VPC] When the “corrected predictions” option was selected in the VPC, the percentiles were corrected but the observations were not.
[VPC] When splitting the VPC by a covariate containing special characters in the categories names, the outlier area was drawn from the middle of the prediction intervals to the empirical percentiles instead of from the border of the prediction interval to the empirical percentiles.
[Correlations between random effects] Decorrelated random effects were missing from the exported charts data.
Monolix / Reporting
[VPC] The “prediction correction” option was not taken into account when generating a report.
Generating a report with no results is now possible, and generating a report with invalid results gives an error, like for PKanalix.
Invalid conditional mean results were not detected while generating a report.
Sycomore
Fixed the resetting of the list of directories when selecting/deselecting.
Simulx
[Import / Export] Exporting a Simulx project to Monolix led to an error if individual parameters in the structural model input={} line were listed on several lines.
[Import / Export] During export from Monolix to Simulx, parameters with different cases (e.g. “Par” and “par”) were considered the same for individual parameters.
[Model] If a model using an analytical solution has a variable depending only on constant regressors and there are occasions, there were issues with the calculations of the analytical solution at occasion change.
[Model] There was a crash when applying a model with an incomplete [INDIVIDUAL] block.
[Simulation] There was a crash when occasions of type “common” were used with one or several simulation elements defined via an external file with a column occ.
[Definition] Elements from external files with a single line were displayed without the id column in “View”.
[Definition] It was not possible to view and add manually regressor values when there were too many of them.
[Definition] There was a crash when using an external file for population parameters with an ID column. An error is now reported.
[Results] Export of output files for the Exploration tab did not work in Simulx2023R1.
[Results] Simulated data were exported to the default location even when clicking Cancel in the pop-up window.
[Plots] The tab Stratify sometimes took a very long time to appear, this has been improved.
[Plots] “Observed data” has been changed to “Simulated data” in plot legend.
Lixoft Connectors – Global
The format of output connectors has been adjusted: “id” is now always lower case, id type is numeric in Simulx and character in Monolix, rep is numeric, group is a factor, original_id is a character. Numeric columns stay numeric even when they include NaNs. Problems of lists inside data.frame were fixed for getCorrelationOfEstimates(), getCAParameterStatistics() and getNCAParameterStatistics(). Treatment washouts from getSimulationResults() are now always numerical 0 or 1 instead of character TRUE or FALSE.
In setData() or newProject(), observationTypes are now optional and the default is continuous. The syntax to define column types has been simplified and ignored columns do not have to be specified.
Various bug fixes for plot connectors.
Lixoft Connectors – PKanalix
NCA results returned as tables and plots by the connectors were not using parameter aliases.
NCA parameter aliases preference were not recognized by setPreferences().
Using setNCAResultsStratification() several times with different filters resulted in a crash.
The filtering by acceptance criteria in setNCAResultsStratification() selected ids with flag = 0 instead of ids with flag = 1.
Partial AUC parameters were missing from the list of computed parameters returned by getNCASettings().
Sometimes the results from getNCAData() were not well updated after loading a new project after another one.
Lixoft Connectors – Monolix
plotVpc() was incorrect when used with TTE data and split by covariates.
Red areas displayed by plotVpc() were incorrect in log scale.
runModelBuilding() with settings gave a warning in R 4.2 and an error in R 4.3.
In case of IOV, getSimulatedRandomEffects() now returns sampled random effects at the ID and OCC levels instead of just the sum.
Mapping arguments were not taken into account in newProject().
The “parameters” argument has been removed from getFixedEffectsByAutoInit().
There was a crash when using setPopulationParameterInformation() with method=”MAP” on a correlation parameter, it now returns an error.
Lixoft Connectors – Simulx
The behavior of getSimulationResults() has been improved when returning large results, and it is now possible to give as argument a list of ids or replicates.
Using setAddLines() on a project without simulation groups resulted in a crash.
The output format of getPopulationElements() and getRegressorElements() in case of distribution has been changed with two columns lowerLimit and upperLimit.
Using defineOutputElement() with an interval larger than 100 gave an error.