RikSchoemaker
commented
6 years ago Hi Ron,
Great to hear you got it to work on all your platforms: congratulations! I cannot comment on your full omega matrix because I've never tried this, but I can confirm that nlme often does not converge with the default settings. What I have done so far is increase the pnlsTol parameter a significant amount. Often 0.01 works, but in this case I needed to ramp it up even further, and I have also increased the tolerance setting for one of the runs. Full syntax:
library(nlmixr)
library(data.table)
#function to add (7) doses to allow SS-type functionality, used in runN025:
SSdata<-function(data=datr,IV=FALSE){
datSS<-data[SS==0]
datSD<-data[SS==1]
setkey(datSD,ID,TIME)
#general solution to allow different times of SS dose and different II values per subject:
datSSD<-datSD[,.(ID,TIME,II)]
nvar<-function(TIME,II){
for(i in seq(1,7)){
r=TIME-(i-1)*II
assign(paste("ret",i,sep=""),r)
}
return(list(r1 = ret1, r2 = ret2,r3=ret3,r4=ret4,r5=ret5,r6=ret6,r7=ret7))
}
#updates datSSD with 7 columns to account for the new dosing times
datSSD[,(paste("V",seq(1,7)-1,sep="")):=nvar(TIME,II)][,TIME:=NULL][,II:=NULL]
index<-melt(datSSD,id.vars="ID",value.name="TIMED")
index[,variable:=NULL]
index<-index[TIMED>0]
setkey(index,ID,TIMED)
datSD2<-merge(datSD,index,by=c("ID"),all=TRUE)
datSD2[,TIME:=TIMED][,TIMED:=NULL]
if (IV==TRUE){
datSDoff<-copy(datSD2)
datSDoff[,TIME:=TIME+AMT/RATE][,AMT:=-1*AMT]
datSD2<-rbind(datSD2,datSDoff)
}
datSS<-rbind(datSS,datSD2)
setkey(datSS,ID,TIME)
dat<-datSS
dat
}
one.compartment.oral.model <- function(){
ini({ # Where initial conditions/variables are specified
# '<-' or '=' defines population parameters
# Simple numeric expressions are supported
lCl <- 1.6 #log Cl (L/hr)
lVc <- log(90) #log V (L)
lKA <- 0.1 #log V (L)
# Bounds may be specified by c(lower, est, upper), like NONMEM:
# Residuals errors are assumed to be population parameters
prop.err <- c(0, 0.2, 1)
# Between subject variability estimates are specified by '~'
# Semicolons are optional
eta.Cl ~ 0.1
eta.Vc ~ 0.1
eta.KA ~ 0.1
})
model({ # Where the model is specified
# The model uses the ini-defined variable names
Cl <- exp(lCl + eta.Cl)
Vc <- exp(lVc + eta.Vc)
KA <- exp(lKA + eta.KA)
# RxODE-style differential equations are supported
d/dt(depot) = -KA*depot;
d/dt(centr) = KA*depot-(Cl/Vc)*centr;
## Concentration is calculated
cp = centr / Vc;
# And is assumed to follow proportional error estimated by prop.err
cp ~ prop(prop.err)
})
}
datr <- read.csv("ORAL_1CPT.csv", header=TRUE,stringsAsFactors=F)
datr$EVID<-ifelse(datr$EVID==1,101,datr$EVID)
datr<-data.table(datr)
datr<-datr[EVID!=2]
runno<-"N023"
data<-datr[SD==1]
fit <- nlmixr(one.compartment.oral.model,data,est="nlme",control = nlmeControl(pnlsTol = .3, tolerance=1e-3))
save(fit,file=paste("fitNLME_ODE_UUI",runno,".Rdata",sep=""))
runno<-"N024"
data<-datr[SD==0]
fit <- nlmixr(one.compartment.oral.model,data,est="nlme",control = nlmeControl(pnlsTol = .1))
save(fit,file=paste("fitNLME_ODE_UUI",runno,".Rdata",sep=""))
runno<-"N026"
data<-datr
fit <- nlmixr(one.compartment.oral.model,data,est="nlme",control = nlmeControl(pnlsTol = .3))
save(fit,file=paste("fitNLME_ODE_UUI",runno,".Rdata",sep=""))
runno<-"N025"
data<-SSdata(datr)
fit <- nlmixr(one.compartment.oral.model,data,est="nlme",control = nlmeControl(pnlsTol = .1))
save(fit,file=paste("fitNLME_ODE_UUI",runno,".Rdata",sep=""))
mattfidler
ronkeizer
wwang-at-github
Hi all,
Great work on the recent developments and release! Got it to work on all platforms now, nice job on that. Basic examples run fine, but encountered a few things when exploring one of the examples a bit deeper, not sure if I'm doing something wrong... Using Oral_1CPT data, and model example as given in help:
Omega matrix used:
Error:
call:
error:
Any ideas on these issues? First one seems some matrix definiteness issue perhaps? But second one occurs also with diagonal block.
Thanks!