Release Notes MonolixSuite2019R1

MonolixSuite 2019R1 Release Notes

April 2019

The document is the release notes for MonolixSuite2019R1 and contains the new software description,  most of the evolution of the software along with the data set and Mlxtran management.

Data set

Enhancement:

  • Addition of an IGNORE LINE column type to allow to ignore the whole ligne. In opposition to IGNORE OBSERVATION that only ignores the observation, IGNORE LINE ignores the entire information of the line (amount and regressor for instance).
  • Occasion values can be larger than 100.

Bug fix:

  • When there are several administration types and SS on several administration, the additional doses were not properly managed. Different dose types are now considered independently of each other and therefore not considered as overlapping.
  • When EVID=4 and SS=1, the doses where not necessarily well added when there was an overlap with previous doses
  • When there is no OCCASION column, the error and warning messages containing “subject-occasions” are replaced by “subject”
  • Identification of implicit washouts (time overlap) in case of several occasion columns

Mlxtran

Enhancement:

  • The following libraries were added
  • Count model library
  • Double absorption PK model library
  • Empty and reset macros now allow to reset a variable of the system to 0 or to its initial value at a given dosing time.
  • Improved implementation of analytical solutions in case of infusion in data set or zero-order absorption. Large time values do not lead to “nan” anymore.

Bug fix:

  • Variables in the “table=” statement are now checked at model compilation. In previous versions, a wrong variable name would result in a freeze at the end of the “population parameters” task.
  • Washouts with EVID=3 were not interrupting infusions administrations.

Datxplore  

Enhancement:

  • The covariate representation has changed. We have split the plot in two parts, the observations in one plot and the covariates in the other one. There is now the possibility to represent all the covariates w.r.t. all the covariates (matrix view) and have the possibility to select any combination of subplots. In addition, a new tab “covariates” presents statistics on covariates.
  • In the menu, a button “Export to Monolix” allows to automatically open Monolix with the data set already set.
  • In the case of TTE data, there is now an “information” button with the total number of subjects and the number and percentage of censored observations.
  • In the case of TTE data, the calculation method of the mean number of event in case of repeated events has been improved to better take into account censoring.
  • It is now possible to save the plots
  • A tab preferences allows to change the layout of the plotting region (label sizes, etc)
  • The number of subjects in categorical covariate groups for stratification is now displayed.
  • It is now possible to type manually the bounds of the continuous covariate groups for stratification.
  • When “individual selection” is used, censored data is displayed using censoring intervals rather than the LOQ.
  • Id is now available as a categorical covariate. It can be used to highlight a specific ID.
  • A preference section in the menu has been added. It allows to specify the data set headers that should be recognized automatically when loading a data set.
  • A Monolix project can now to loaded into Datxplore

Bug fix:

  • Issues with combined stratification groups have been fixed.

Monolix

Monolix Interface

Data tab

  • Graphical filters for each column by selecting categories or value ranges
  • Data from a subject can be hidden
  • The choice of the type of observation (continuous/categorical-count/event) has been moved to a clearer place
  • Additional doses can be visualized as additional rows colored in blue, and ignored lines (with the column-type IGNORED LINE or because of missing ID or OCCASION) can be colored in pink or can be hidden
  • Highlighted id in the plots is also highlighted in the data, and the page containing this id is automatically shown

Structural model tab

  • Addition of two new libraries (count and PK double absorption)
  • A new panel has been added that allows to manually map model outputs to data observation ids, thus giving more flexibility when several model outputs and/or several observation types are present.
  • When selecting a model from the library with a parameter F, F1 or Imax, the parameter distribution is automatically set to logitnormal and the initial value to 0.5.
  • When changing the structural model file from a library model to another model from the same library, the filter choices are remembered.

Initial estimates tab

  • An “auto init” button was added in the check initial estimates part when the structural model comes from the PK library. It automatically proposes good initial values.
  • It is possible to add a reference (i.e freeze the current prediction), to update and remove it to be able to compare several initial estimates.
  • When clicking on “use last estimates”, initial values for the correlations are also set. Custom initial values for the correlation cannot be set via the interface but can be set by editing manually the mlxtran project file.

Statistical model and tasks tab

Observation model

  • The display of the observation models was updated to be more consistent with the individual model and simplify the representation of multiple outputs models.

Individual model

  • In the covariates information, the weighted mean information was added.
  • A button to directly transform a continuous covariate into its log transformed value centered by the weighted mean has been added next to each continuous covariate.
  • Logit distributions can be between any bounds (not necessarily 0 and 1).
  • Bug fix: results were not reloaded when the bounds of a logit transformation of the observations were not positive integers.

Comment tab

A new “Comment” tab allows to write text about the project with markup and markdown formatting.

Results

General

  • All tables in the results tab are now copyable to the clipboard. When pasted in Excel or Word, the layout is kept.

Individual parameters

  • In addition to the summary of the conditional modes and the conditional means, all the individual parameters are displayed along with the covariates. In addition, it is possible to sort the table.

Tests

  • Evolution of the method to take correlations between replicates into account in a more robust way
  • Covariates and correlations present in the current model are highlighted in blue.
  • The tests of the random effects are now split by level of variability (IIV, IOV, …) to propose a better diagnosis of the inter-occasion-variability.
  • The tests are now done after the task Conditional Distribution rather than after the task Plots.
  • Test for symmetry has been changed from Van Der Waerden to Miao-Gel-Gastwirth

Likelihood

  • A corrected BIC was added. Compared to the BIC, it has better asymptotic properties by weighting differently parameters with and without variability.

Proposal

  • A new section “proposal” was added. In this section, automatic proposals of improvements for the statistical model are presented, based on comparisons of many correlation, covariate and error models. Model selections are performed by using a BIC criteria based on the current individual parameters drawn from the conditional distribution.

Convergence assessment

  • The convergence assessment was moved from a pop-up to a dedicated tab.
  • There are now buttons to manually save figures to png or svg.
  • When loading a project, the convergence assessment results are also reloaded.

Bug fix:

  • Intervals for sensitivity on initial parameters are no longer centered around the estimated values when there are results. They are centered around the initial values.
  • Crash if SAEM was not computed due to NaNs has been fixed

Monolix calculation engine

Parameter estimation

  • Optimization of parameters without inter-individual variability has been improved for the “no-variability” method (parameters are optimized one after the other instead of simultaneously), for the “decreasing variability” method (variability decreasing rate is better updated for smoothing phase when stopping exploratory phase before the max number of iterations), and for the “variability at the first stage” method (additional checks for unusual cases).
  • A new setting simulatedAnnealingIterations can be used to disable the simulated annealing after a given number of iterations. This setting does not appear in the interface but can be added directly to the mlxtran project file.
  • In case of IOV and independent occasions (washout), MCMC has been improved to be faster.

Bug fix:

  • The case where the decreasing rate in the simulated annealing is greater than 1 (the variance can not increase too fast) is now properly handled.
  • Bug fixed in estimation of beta when there are latent covariates and the beta parameter is fixed

Fisher Information Matrix calculation

  • In case of stochastic approximation, several code optimizations (memory allocation, multi-thread, keep values in memory) have been made to speed up the calculations.
  • In case of stochastic approximation, the methodology was updated to test several combinations of the set of monte carlo matrices and keep the best one.

Bug fix:

  • In case of linearization, when there are several and different censored observations per individual, a bug has been fixed.
  • In case of IOV, reproducibility of results could not be ensured in some rare cases

Likelihood

  • Improved error message when too many NaNs prevent the calculation by importance sampling

Monolix model building

There is a new tab to build automatically the statistical model. Three algorithms are proposed.

  • SCM proposes the classical Stepwise Covariate Modeling algorithm.
  • COSSAC (COnditional Sampling use for Stepwise Approach based on Correlation tests) makes use of the information contained in the base model run to choose which covariate to try first.
  • SAMBA (Stochastic Approximation for Model Building Algorithm) is an iterative procedure that identifies at each step how to best improve the statistical model components (residual error model, covariate effects, correlations between random effects).

Monolix plots

Evolution of the calculation of the simulated BLQ

In previous versions, the value of simulated BLQ was the same for all individual parameters (conditional mean, conditional mode, simulated parameters), leading to a mismatch between the simulated BLQ and the corresponding prediction (in Obs versus pred or the IWRES for instance). We now match the simulated BLQ with the prediction that has been used to generate the simulated BLQ.

Global

  • When highlighting an individual in a plot, the same individual is highlighted in all other plots and in the Data tab.
  • Id is now available as a categorical covariate. When clicking on an individual ID in the stratify tab, it is highlighted in all plots
  • The count of individuals by modalities for categorical covariates is now displayed in the Stratify panel.
  • The bounds of the groups for continuous covariates can be changed manually.
  • The calculation of the latent covariate category to which each individual belongs has been improved

Observed data chart

  • In the case of TTE data, there is now an “information” button with the total number of subjects and the number and percentage of censored observations.
  • In the case of TTE data, the calculation method of the mean number of event in case of repeated events has been updated.

Observation versus prediction chart

  • There are now the same limits on both axis.

Scatterplot of the residuals

  • When the selected individual estimates are from the conditional distributions, the individual predictions on the x-axis are also based on the individual parameters from the conditional distribution.
  • For the IWRES using individual parameters from the conditional distribution, the confidence interval calculation now takes into account the fact that replicates may be correlated

Distribution of the random effects chart

  • In case of IOV, it is possible to split the random effects by level.

Correlation of the random effects chart

  • In case of IOV, it is possible to split the random effects by level.
  • It is possible to display decorrelated random effects.

Prediction distribution chart

  • Bug fix: The tooltip of the ID was missing on the censored data.

Visual predictive checks chart

  • For TTE data, the calculation method of the mean number of event in case of repeated events has been updated.
  • For TTE data, a new methodology (Turnbull intervals) which prevents the survival curve to fall to 0 has been implemented.

Likelihood contribution:

  • When more than 50 individuals, individuals are grouped into groups of similar likelihood

Monolix project definition, settings, and outputs

  • When there is IOV, the value by levels are added into the file simulatedRandomEffects.txt
  • New setting SimulatedAnnealingIterations to define the number of iterations for which the simulated annealing option is applied.
  • New option in the menu preferences: It is possible to define new automatically recognized header names.
  • It is possible to import a PKanalix or Datxplore project. This will set the data set information.
  • Bug fix: Projects for which the model file is located on another disk now properly reload.

Monolix Connectors

  • The following connectors have been added: computeChartsData, getProjectInformation, runModelBuilding, getModelBuildingResults, getModelBuildingSettings, stopModelBuilding, getLixoftConnectorsState, setIndividualLimits

Simulx

  • Update of the communication technology between R and the C++ code for Simulx
  • It is not mandatory anymore to have at least one individual with each categorical covariate category defined in the model.

PKanalix

We provide a new software for NCA and CA analysis.

Sycomore

We provide a new software for Monolix project managements.

General

  • Installation: Silent installation with config file
  • Improved handling of username with special characters

Jonathan CHAUVINRelease Notes MonolixSuite2019R1