When creating a controller ggPMX is dropping large portion of the data: ggPMX 1.2.8 (but also earlier versions)

[biencan1]

Example 1: ER model, only response variable fitted. DV values contain zero

1) plot generated using ggPMX function: ctr %>% pmx_plot_npde_time

2) plot recreated by reading in data from controller

dat2plot<-ctr$data$predictions

ggplot(dat2plot,aes(x=TIME,y=NPDE)) +
  geom_point()+
  geom_smooth(se = FALSE,color="red")+
  theme_bw()

3) plot recreated by reading in directly Monolix outputs plot source data:

ordat2plot<-read.csv(
  file.path(model.dir,model,'ChartsData/ScatterPlotOfTheResiduals/LIDV_residuals.txt'),sep=','
)

ggplot(ordat2plot,aes(x=time,y=npde)) +
  geom_point()+
  geom_smooth(se = FALSE,color="red")+
  theme_bw()

Example 2: PopPK model, dataset contains censored values

1) plot generated using ggPMX function:

ctr %>% pmx_plot_npde_time

2) plot recreated by reading in data from controller

dat2plot<-ctr$data$predictions

ggplot(dat2plot,aes(x=TIME,y=NPDE)) +
  geom_point()+
  geom_smooth(se = FALSE,color="red")+
  theme_bw()

3) plot recreated by reading in directly Monolix outputs plot source data

ordat2plot<-read.csv(file.path(model.dir,model,'ChartsData/ScatterPlotOfTheResiduals/LIDV_residuals.txt'),sep=',')

ggplot(ordat2plot,aes(x=time,y=npde)) +
  geom_point()+
  geom_smooth(se = FALSE,color="red")+
  theme_bw()

Dropping of zero or BLOQ values doesn’t seem to be the source of the problem: different fill for BLOQ data

ggplot(ordat2plot,aes(x=time,y=npde)) +
  geom_point(aes(color=factor(censored)))+
  scale_colour_manual(values = c("0"="black","1"="blue"))+
  geom_smooth(se = FALSE,color="red")+
  theme_bw()

filter out censored data

ggplot(filter(ordat2plot,censored==0),aes(x=time,y=npde)) +
  geom_point()+
  geom_smooth(se = FALSE,color="red")+
  theme_bw()

replot using simulated BLOQ

ggplot(ordat2plot,aes(x=time,y=npde_simBlq)) +
  geom_point(aes(color=factor(censored)))+
  scale_colour_manual(values = c("0"="black","1"="blue"))+
  geom_smooth(se = FALSE,color="red")+
  theme_bw()

[biencan1]

Feedback from Lixoft: rounding all columns of the dataset to 6 significant digits before using it in Monolix will solve the issue. In the next version, we will increase the number of significant digits in the output to resolve this problem.

[biencan1]

After testing the work around of rounding to 6th significant digit seems to have resolved the issue. See snapshot below as example.

Original dataset:

ctr$data$predictions created from this model fit (some time points dropped):

ctr$data$predictions created from a run with a modified data set where rounding was applied to time variable using mutate(TIME=signif(TIME,6))

[tynsci]

@biencan1

• closing this since it is an issue with Monolix
• feel free to reopen if you think something should be done with ggPMX at this time to address the problem