Evolutions of the Mlxtran language now allow to share a model in the workflow from model exploration (using Mlxplore), to parameter estimation (using Monolix), and simulation (using Simulx). Therefore, the proposed evolutions of the language merge the needs of all these applications and the model written with this new syntax will be compatible with all Lixoft software. See here to see the evolution and the eventual impact on your model.
Main improvements and new features:
– The interface was redesigned for clarity and simplicity.
– A button providing the statistical model summary was added
– A button providing a direct browse to the demo was added
– All the libraries are now using the Mlxtran syntax. Therefore, if you were using a model from the Matlab library, it will automatically be transformed into its Mlxtran equivalent model when saving. The Matlab models are still available although their use is no longer recommended.
Enhancements:
Numerical
– Improved LL computation when the sensitivity analysis is verified
– Improved simulated annealing initialization
System
– Added configuration option for sysadmin to enforce number of structural model threads and maximum computational matlab threads
Naming
– Latent covariate by default are now called lcat, lcat1, etc
– Error model parameters are by default now called (a,b,c) if there is only one continuous error model
– The autocorrelation parameter is by default now called (r) if there is only one output with autocorrelation
– IOV variability parameters are by default now called (gamma_) if there is only one level, (gamma1_, gamma2_) if there are more
– Latent covariate probabilities are by default now called (plcat_1, plcat_2) for a latent covariable with three categories if there is only one latent covariate, (plcat1_1, plcat1_2, ) if there are more.
Bugs fixed:
– bug fix: fixed effects were not correctly initialized when they were not simulated
– bug fix: fixed problems estimating autocorrelation (SAEM) and the corresponding Fisher information matrix
– bug fix: limits of the confidence intervals in VPC
– bug fix: dde solver precision in interpolation of delays
– bug fix: computation of the log-likelihood of a multivariate gaussian vector on an interval during FIM computation by linearization for censored data was corrected
– bug fix: simulated annealing on standard error parameters for mixture projects was not well used
– bug fix: when SS and ADDL were used at the same time in the data set, it was not well interpreted.
– bug fix: occasion time management was not correct when evid=1 and the first occasion of the subject had no observations
– bug fix: corrections when several covariates are created with the same initial covariate
– bug fix: corrections when IOV occurs with discrete models
– bug fix: correction on individual parameter estimate in conditional mode with no IIV
