MonolixSuite 2018R2 Release Notes
July 27th, 2018
The document is the release notes for MonolixSuite2018R2 and contains the bug fix from MonolixSuite2018R1 along with the data set and Mlxtran management.
Data
* There was a bad rejection of CMT data column header even if already recognized as IGNORE, if the default header recognition was directly accepted.
* SS rules for EVID=4 and SS=1 changed, the SS doses are now always added.
* ‘c’ is no more recognized as alias for covariate.
* The number of obsid in the “Observation” column is now updated after changing a column from “ignored observation”to “ignore”.
* When there is no time nor regressor columns in data set, the split by subject was not correctly done.
Monolix project
* [load] error messages for using covariates that are invalid in the data set were improved (e.g. only one category).
* [load] The reload of the number simulated parameters setting for conditional distribution task after a an error of loading a bad project is now correctly done.
* [load] The summary of estimated conditional means was not correctly reloaded.
* [Export / save] Initial estimates in mlxtran project file are now saved at 1e-16 precision. Output files used at reload are saved at 1e-16. Output files not used at reload at 1e-6.
* [Export / save] Default export path uses now the complete base name of the saved project (eg project.1 for project.1.mlxtran).
* [Export / Simulx] Fix wrong observation order w.r.t. ytypes for simulx (monolix2simulx).
* [Preferences] The project is not tagged as changed (star in project name) anymore when the preferences settings are changed.
* [Log] Errors, warnings & information raised during import / export tasks (e.g when reading output files at reload or when writing of the output is not possible) are now notified to the gui.
* [load] All observation types from the data set are available for future use. Those missing in “[CONTENT]” get a default name and are assumed to have continuous values. It eases switching for a model with more outputs than the current one.
* [export] Predictions are now correctly written in predictions.txt file for bsmm models
* [export] Fix bad iwres values in predictions.txt when error model distribution is not normal
Model
* [Model library] In the PK and PKPD libraries with joint oral/extravascular and bolus/infusion administration, models with the bioavailability were added.
* [Model library] In the TMDD library, models with transit compartments were added.
* [FIT] The fit between observations and outputs in the case of several discrete observation models is now correctly done.
* [Model library] some PKPD file names were badly parsed resulting in a wrong filtering
Initial estimates
* [check initial estimates] Ids with no measurement for an input were displayed with an empty plot. They are now removed.
* [check initial estimates] Pagination was not working after changing the output
GUI
* New & load in project menu entries are now disabled during scenario and assessment runs
* Initial estimates tab is now locked during scenario and assessment runs
* Fixed parameters are now displayed as read-only values in assessment initialization dialog
* Maximum number of iterations and tolerance for parameters without variability can now be changed in the settings of SAEM.
Algos
* [SAEM] There is no more crash after SAEM when there are several observation models with bsmm
* [SAEM] IOV: fix if some parameters have no variability on ID level and no covariates are used
* [SAEM] IOV: fix if some parameters have no variability at all, option “variability at the first stage” is selected in pop. parameters settings and none of the parameters have variability at the last level (e.g column occ tagged as occasions but no iov defined)
* [SAEM] IOV: fix if a variability level which is not the last one is empty (e.g none of the parameters have iiv)
* [SAEM] IOV: fix if some parameters have no inter-individual variability, no parameter have variability at the last level, and artificial variability is selected in the settings of saem: sometimes the artificial variability could reach very small values that caused a parameter switch at 80% of the maximum number of iterations of the exploratory phase.
* [SAEM] Simulated censored method was updated to better manage extreme values
* [All] Simulated censored are now different for each chain. This stabilizes convergence in case of constant error models.
* [Loglikelihood] IOV: fix in importance sampling when the last level of variance is not full (e.g some parameters do not have iov)
* [Loglikelihood] Fix LogLikelihood calculation if there are several chains and bayesian priors.
* [MPI] Threads option is distributed for monolix mpi
* [Results / Std Errors] Latent probabilities are not taken into account for the eigenvalues anymore
* [Results/Tests] Robustness of the statistical tests was improved.
* [Results/Tests] Computing the ANOVA test for a categorical covariate no longer crashes when it counts as many modalities as individuals.
* [Results/Tests] Security for forbidden values in statistical test (Pearson)
* [FISHER] Computing the Fisher Information Matrix no longer crashes if there is only parameters for latent covariate to estimate.
* [FISHER] Removal of a wrong warnings in stochastic approximation if there is nan in derivative of parameters which is fixed.
Plots
* [All] When running only one plot, the display of the progress bar is now displayed.
* [All] gamma unicode character is now supported on windows.
* [All] When there is an error in the plots generation (nan are generated for example), the error message is now provided to the user in the Notifications.
* [All] Points are now excluded from log scale compared to epsilon instead of zero. This prevents all points being squeezed on on x or y axis.
* [All] Empty charts are no longer taken into account when computing common domains (for “same limits” settings)
* [All] The gap between ticks and axis is now updated after changing the ticks font size.
* [output plot] Individuals with no observations are no longer taken into account in information dialog.
* [individual fits] When there is no variability at all, the individual fits were not well plotted (black windows)
* [individual fits] The time grid was badly defined when occasions are not joined for subjects with only one observation.
* [individual fits] With several observation models and IOV, if one model has subject occasions without observation nor regressor but doses, the concatenation of the occasion was badly done.
* [residuals] Censored data were used in bins computation even when the corresponding toggle was set to ‘false’, thus the bins were not updated after ‘useCensoredData’ toggle change.
* [residuals] In bsmm case, the IWRES computation is now updated after SAEM and/or IndivEstim.
* [residuals] fix computation in bsmm case
* [residuals] npde computation : improve rank definition when pwres and several simulation residuals are equal (take the middle position instead of the upper_bound (mlxsuite2018R1) or of the lower_bound (mlxsuite2016))
* [Random effects] Distributions of the random effects with latent covariate are now correct
* [All] Wrong observation values for bsmm models after saem were fixed
* [predictive checks] Prediction interval building does not crash anymore when stratification leads to empty subplot
* [charts export] When there are prediction intervals on a splitted figure, all prediction intervals of all subplots are corrected exported to SVG figure.
* [charts export] We now embed the font used in the GUI when exporting figures.
* [charts export] We now export legend (in chart image file or in a separate file)
* [Export charts data / Observation vs Predictions] There was an extra randomly filled column in xxx_visualGuides.txt.
* [Export charts data / Distribution of the residuals]: in charts data table, columns for iwRes have the correct length when using “conditional distribution” and there are several values per individual.
* [Bins] do not try to fit maximumNbOfPoints per bin condition when the number of bins is fixed
* [Bins] Fix in computation function. Prevents from crashing if the data points are too similar (e.g in scatter plot of residuals)
* Use last saved settings after a ‘save as’ in new plots
* [random effects] selection of individual rows/columns with layout button now displays the correct subplots
* [individualFits] fix in occasion merge when some subject occasions have only doses
Stratify
* Automatic continuous groups sizing now prevents them from being of null length
MlxConnectors
* initializeMlxConnectors(): improved error message when the dll is not found
* manage multi-dimensional arrays in .encodeNumericals()
* unload mlxConnectors library when initialization fails. This prevents from inconsistent behaviour when the user kills licenseActivate window
* Typo corrections
* computePrediction(): 1st column of individualParameters input dataFrame is not suppressed anymore. This was leading to wrong output values for computePrediction.
* newProject(): header types passed in argument are taken into consideration before data validation. This avoids errors when the default tagging as the data set is not ok.
Mlxtran
* There was a bad reset with EVID=3
* if t0 is defined (ODE starting again from t0 instead of t)
* for a TTE or continuous Markov chain (lost history of elapsed time)
* There was an issue in simulations for TTE models with regressors
* The analytical solutions did not use IOV parameter values, even with washouts or new time periods
* There were instability of TTE graphic with IOV if an occasion has no simulated event
* Accept very large integers, converting to double-precision floating point values. This allows a safe save-reload of very large double-precision values without decimals.
Mlxplore
* We now take into account the new saved project in mlxplore recent files (from a monolix project)
* use the time grid from the <OUTPUT> section even if there is a <DATAFILE> without a <DESIGN>. There will be no doses.
Mlxeditor
- The “bsmm” keyword is now well recognized
