333 lines
11 KiB
R
333 lines
11 KiB
R
|
library(tidyverse)
|
||
|
|
library(gslnls)
|
||
|
|
library(ggplot2)
|
||
|
|
library(reshape2)
|
||
|
|
library(drc)
|
||
|
|
library(ggridges)
|
||
|
|
|
||
|
|
library(car)
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
# Pseudoreplication
|
||
|
|
# simulation of replication strategies
|
||
|
|
|
||
|
|
Calc_DilRes <- function(as = 10, bs = 1, cs = -6, ds = 110, at = 10, bt = 1, r=0.01,
|
||
|
|
ct = cs, dt = 110, sd_fac=0.1, gt=1, gs=1, log_concREF, log_concTEST, # -r
|
||
|
|
THEOconc ,
|
||
|
|
heteroNoise=FALSE, noDilSeries, noDils) {
|
||
|
|
|
||
|
|
# browser()
|
||
|
|
yAxfac <- (as-ds)
|
||
|
|
log_cREF_l <- melt(log_concREF)
|
||
|
|
log_doseREF <- log_cREF_l$value
|
||
|
|
log_cTEST_l <- melt(log_concTEST)
|
||
|
|
log_doseTEST <- log_cTEST_l$value
|
||
|
|
isRef <- rep(c(1,0),1,each=length(log_concREF))
|
||
|
|
isSample <- rep(c(0,1),1,each=length(log_concTEST))
|
||
|
|
#um with equal sample without
|
||
|
|
ro_jit <- as*isRef + at*isSample + (ds*isRef + dt*isSample - as*isRef - at*isSample)/
|
||
|
|
(1 + isRef*exp(bs*(cs - log_doseREF)) + isSample*exp(bt*(ct -log_doseTEST)))
|
||
|
|
|
||
|
|
ro_jit_ <- abs(ro_jit)
|
||
|
|
all_var <- rbind(log_cREF_l, log_cTEST_l)
|
||
|
|
|
||
|
|
ro_jit2 <- cbind(all_var, ro_jit_, rep(THEOconc, 6), isRef, isSample)
|
||
|
|
# if (noDilSeries==3) {
|
||
|
|
# colnames(ro_jit2) <- c("R_dil1","R_dil2","R_dil3","T_dil1","T_dil2","T_dil3", "log_dose")
|
||
|
|
# } else if (noDilSeries==2) {
|
||
|
|
# colnames(ro_jit2) <- c("R_dil1","R_dil2","T_dil1","T_dil2", "log_dose")
|
||
|
|
# }
|
||
|
|
|
||
|
|
colnames(ro_jit2) <- c("dil_no","repl","trueLogConc" ,"readout","log_dose","isRef","isSample")
|
||
|
|
return(ro_jit2)
|
||
|
|
}
|
||
|
|
|
||
|
|
|
||
|
|
RestrMF <- function(data, M) {
|
||
|
|
|
||
|
|
all_l <- data
|
||
|
|
#plot(all_l$log_dose, all_l$readout, col=all_l$sample)
|
||
|
|
CORro <- cor(all_l$log_dose[1:8], all_l$readout[1:8])
|
||
|
|
if (CORro<0) B <- -1 else B <- 1
|
||
|
|
|
||
|
|
# browser()
|
||
|
|
startlist <- list(a = min(all_l$readout), b = B, cs=mean(all_l$log_dose),
|
||
|
|
r=0.0001, d = max(all_l$readout))
|
||
|
|
mr <- tryCatch({
|
||
|
|
gsl_nls(fn = readout ~ a + (d - a)/(1 + exp(b*((cs-r*isSample)-log_dose))),
|
||
|
|
data = all_l,
|
||
|
|
start = startlist, #trace=T,
|
||
|
|
control = gsl_nls_control(xtol = 1e-6, ftol = 1e-6, gtol = 1e-6))
|
||
|
|
}, error = function(msg){
|
||
|
|
return(0)
|
||
|
|
} )
|
||
|
|
|
||
|
|
if (length(mr)>1) {
|
||
|
|
s_mr_coef <- summary(mr)$coefficients
|
||
|
|
s_mr <- summary(mr)
|
||
|
|
} else { s_mr_coef <- rep(0,4); s_mr <- 0 }
|
||
|
|
|
||
|
|
if (M == "R") {
|
||
|
|
return(s_mr_coef)
|
||
|
|
} else if (M =="CI") {
|
||
|
|
if (length(mr)==1) return(rep(0,4))
|
||
|
|
(RelPot_mr <- exp(s_mr_coef[4,1])) # r exponentiated
|
||
|
|
|
||
|
|
POTr_CI <- exp(confint(mr, "r", method = "asymptotic", level=0.95))
|
||
|
|
# Pure error
|
||
|
|
#browser()
|
||
|
|
suppressWarnings(FitAnova <- anova(lm(readout ~ factor(log_dose)*isSample, all_l)))
|
||
|
|
meanPureErr <- FitAnova[4,3]
|
||
|
|
SEsPure <- sqrt(diag(vcov(mr)/s_mr$sigma^2)*meanPureErr)
|
||
|
|
|
||
|
|
# PLA software: 1.0252 - 1.4242
|
||
|
|
lCIpure <- exp(s_mr_coef[4,1]-qt(0.975,nrow(all_l)-16)*SEsPure[4])
|
||
|
|
uCIpure <- exp(s_mr_coef[4,1]+qt(0.975,nrow(all_l)-16)*SEsPure[4])
|
||
|
|
if (is.infinite(uCIpure)) uCIpure <- NA
|
||
|
|
if (is.infinite(POTr_CI[2])) POTr_CI[2] <- NA
|
||
|
|
relCIpure <- c(lCIpure/RelPot_mr, uCIpure/RelPot_mr)
|
||
|
|
CIsDF <- c(RelPot_mr,POTr_CI, lCIpure, uCIpure)
|
||
|
|
names(CIsDF) <- c("potency","lCI","uCI","lpureCI","upureCI")
|
||
|
|
return(CIsDF)
|
||
|
|
} else if (M=="U") {
|
||
|
|
if (length(mr)==1) return(rep(0,4))
|
||
|
|
|
||
|
|
startlstmu <- list(as = min(all_l$readout), bs = B, cs = mean(all_l$log_dose), ds = max(all_l$readout),
|
||
|
|
at = min(all_l$readout), bt = B, ct = mean(all_l$log_dose), dt = max(all_l$readout))
|
||
|
|
mu <- tryCatch({
|
||
|
|
gsl_nls(readout ~ as*isRef + at*isSample + (ds*isRef + dt*isSample - as*isRef - at*isSample)/
|
||
|
|
(1 + isRef*exp(bs*(cs -log_dose)) + isSample*exp(bt*(ct -log_dose))),
|
||
|
|
data = all_l,
|
||
|
|
start = startlstmu, trace=T,
|
||
|
|
control = gsl_nls_control(xtol = 1e-10, ftol = 1e-10, gtol = 1e-10))
|
||
|
|
}, error = function(msg){
|
||
|
|
return(0)
|
||
|
|
} )
|
||
|
|
if (length(mu)>1) {
|
||
|
|
smu_coef <- tryCatch({
|
||
|
|
summary(mu)$coefficients
|
||
|
|
}, error = function(msg){
|
||
|
|
return(rep(NA, 8))
|
||
|
|
} )
|
||
|
|
|
||
|
|
#s_mu <- summary(mu)
|
||
|
|
return(smu_coef)
|
||
|
|
} else return(rep(0,8))
|
||
|
|
|
||
|
|
} else return()
|
||
|
|
}
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
Dil8F <- function(Cn1) {
|
||
|
|
CSer <- matrix(NA, nrow=8, ncol=3)
|
||
|
|
for (i in 1:8) {
|
||
|
|
Cn1 <- Cn1/dilfactor+rnorm(3,0,SD*Cn1)
|
||
|
|
CSer[i,] <- Cn1
|
||
|
|
}
|
||
|
|
return(CSer)
|
||
|
|
}
|
||
|
|
|
||
|
|
DilDirF <- function(OrigMenge) {
|
||
|
|
CDir <- matrix(NA, nrow=10, ncol=3)
|
||
|
|
for (DilExp in 1:10) {
|
||
|
|
C_d <- OrigMenge/dilfactor^DilExp+rnorm(3,0,SD*OrigMenge/dilfactor^DilExp)
|
||
|
|
CDir[DilExp,] <- C_d
|
||
|
|
}
|
||
|
|
return(CDir)
|
||
|
|
}
|
||
|
|
|
||
|
|
homhetCIF <- function(AssD_) {
|
||
|
|
# Confidence level
|
||
|
|
alpha <- 0.05
|
||
|
|
z <- qnorm(1 - alpha/2) # 1.96 for 95% CI
|
||
|
|
|
||
|
|
### 1. Homogeneously Weighted CI ###
|
||
|
|
mean_logRP <- mean(AssD_$logRP)
|
||
|
|
var_mean <- mean(AssD_$SE^2) / nrow(AssD_)
|
||
|
|
SE_mean <- sqrt(var_mean)
|
||
|
|
|
||
|
|
CI_homo <- c(mean_logRP - z * SE_mean, mean_logRP + z * SE_mean)
|
||
|
|
CI_homo_exp <- exp(CI_homo)
|
||
|
|
# cat("Homogeneous weighting:\n")
|
||
|
|
# cat(round(exp(mean_logRP),3), "95% CI =", round(CI_homo_exp[1], 3), "-", round(CI_homo_exp[2], 3), "\n\n")
|
||
|
|
|
||
|
|
### 2. Heterogeneously Weighted CI ###
|
||
|
|
weights <- 1 / (AssD_$SE^2)
|
||
|
|
weighted_mean <- sum(weights * AssD_$logRP) / sum(weights)
|
||
|
|
var_weighted <- 1 / sum(weights)
|
||
|
|
SE_weighted <- sqrt(var_weighted)
|
||
|
|
|
||
|
|
CI_hetero <- c(weighted_mean - z * SE_weighted, weighted_mean + z * SE_weighted)
|
||
|
|
CI_hetero_exp <- exp(CI_hetero)
|
||
|
|
res <- c(exp(mean_logRP) , CI_homo_exp, exp(weighted_mean), CI_hetero_exp )
|
||
|
|
return(res)
|
||
|
|
|
||
|
|
# cat("Heterogeneous weighting:\n")
|
||
|
|
# cat("Combined RP =", round(exp(weighted_mean), 3), "\n")
|
||
|
|
# cat("95% CI =", round(CI_hetero_exp[1], 6), "-", round(CI_hetero_exp[2], 6), "\n")
|
||
|
|
}
|
||
|
|
|
||
|
|
Conc <- c(1)
|
||
|
|
for (x in 1:10) Conc <- c(Conc, (1/3^x))
|
||
|
|
|
||
|
|
|
||
|
|
N <- 10000
|
||
|
|
OrigMenge <- 1
|
||
|
|
dilfactor <- 3
|
||
|
|
SD <- 0.02
|
||
|
|
|
||
|
|
rm(CIsM_all, DevTargC3_)
|
||
|
|
for (x in 1:3) {
|
||
|
|
if (x==1) { Repl1 <- x; Repl2 <- x } # Anzahl Replikate für C2 und C3
|
||
|
|
if (x==2) { Repl1 <- 1; Repl2 <- 3 }
|
||
|
|
if (x==3) { Repl1 <- x; Repl2 <- x }
|
||
|
|
|
||
|
|
Flags <- vector()
|
||
|
|
SDs1 <- SDs2 <- POT <- DevTargC3 <- hhCIV <- c()
|
||
|
|
for (nTrials in 1:N) {
|
||
|
|
C2 <- OrigMenge/dilfactor+rnorm(Repl1,0,SD*OrigMenge)
|
||
|
|
C3 <- C2/dilfactor+rnorm(Repl1,0,C2*SD)
|
||
|
|
CSer <- Dil8F(C3)
|
||
|
|
|
||
|
|
# "true" replication
|
||
|
|
C2_1 <- OrigMenge/dilfactor+rnorm(Repl2,0,SD*OrigMenge)
|
||
|
|
C3_1 <- C2_1/dilfactor+rnorm(Repl2,0,SD*C2_1)
|
||
|
|
CSer_3 <-Dil8F(C3_1)
|
||
|
|
|
||
|
|
|
||
|
|
# Flag <- sd(CSer)<sd(CSer_3)
|
||
|
|
# SDs1 <- c(SDs1, sd(CSer))
|
||
|
|
# SDs2 <- c(SDs2, sd(CSer_3))
|
||
|
|
#Flags <- c(Flags, Flag)
|
||
|
|
|
||
|
|
# all from stem solution directly
|
||
|
|
CDir <- DilDirF(OrigMenge)
|
||
|
|
CDir2 <- DilDirF(OrigMenge)
|
||
|
|
if (x==1) {
|
||
|
|
DevTargC3 <- c(DevTargC3, mean(C3), mean(CDir[2,]))
|
||
|
|
log_concREF <- log(CSer)
|
||
|
|
log_concTEST <- log(CDir[3:10,])
|
||
|
|
testC <- "pseudo"; refC <- "direct"
|
||
|
|
} # Anzahl Replikate für C2 und C3
|
||
|
|
if (x==2) {
|
||
|
|
DevTargC3 <- c(DevTargC3, mean(C3_1), mean(CDir[2,]))
|
||
|
|
log_concREF <- log(CSer_3)
|
||
|
|
log_concTEST <- log(CDir[3:10,])
|
||
|
|
testC <- "true"; refC <- "direct"
|
||
|
|
}
|
||
|
|
if (x==3) {
|
||
|
|
DevTargC3 <- c(DevTargC3, mean(CDir[2,]), mean(CDir2[2,]))
|
||
|
|
log_concREF <- log(CDir[3:10,])
|
||
|
|
log_concTEST <- log(CDir2[3:10,])
|
||
|
|
testC <- "true"; refC <- "true"
|
||
|
|
}
|
||
|
|
# DevTargC3 <- c(DevTargC3, mean(C3), mean(CDir[2,])) # c(DevTargC3, mean(CSer[4,]), mean(CSer_3[4,]))
|
||
|
|
|
||
|
|
ro_new <- Calc_DilRes( log_concREF=log_concREF,
|
||
|
|
log_concTEST=log_concTEST,THEOconc = log(Conc)[4:11])
|
||
|
|
CIs <- RestrMF(ro_new, M="CI")
|
||
|
|
RM <- RestrMF(ro_new, M="R")
|
||
|
|
|
||
|
|
POT <- c(POT, CIs)
|
||
|
|
|
||
|
|
CIsV <- c()
|
||
|
|
for (z in 0:2) {
|
||
|
|
CL <- RestrMF(ro_new[c((z*8+1):(8*(z+1)),(z*8+25):(8*(z+1)+24)),], M="CI")
|
||
|
|
#print(RestrMF(ro_new[c((z*8+1):(8*(z+1)),(z*8+25):(8*(z+1)+24)),], M="R"))
|
||
|
|
CIsV <- c(CIsV, CL[1:3])
|
||
|
|
}
|
||
|
|
AssD <- t(matrix(CIsV, nrow=3))
|
||
|
|
logAssD <- log(AssD)
|
||
|
|
AssD_ <- as.data.frame(AssD)
|
||
|
|
colnames(AssD_) <- c("RP", "lCI", "uCI")
|
||
|
|
AssD_$SE <- (logAssD[,3]-logAssD[,1])/qt(0.975,11)
|
||
|
|
AssD_$logRP <- logAssD[,1]
|
||
|
|
|
||
|
|
hhCIs <- homhetCIF(AssD_)
|
||
|
|
hhCIV <- c(hhCIV, hhCIs)
|
||
|
|
# p <- ggplot(ro_new, aes(x=trueLogConc, y=readout, col=as.factor(isRef))) +
|
||
|
|
# geom_point() +
|
||
|
|
# geom_line() +
|
||
|
|
# ggtitle(paste(nTrials, x)) +
|
||
|
|
# theme_bw()
|
||
|
|
# print(p)
|
||
|
|
#
|
||
|
|
# p2 <- ggplot(ro_new, aes(x=log_dose, y=readout, col=as.factor(isRef))) +
|
||
|
|
# geom_point(size=4) +
|
||
|
|
# ggtitle(paste(nTrials, x)) +
|
||
|
|
# geom_vline(xintercept = c(RM[3,1],(RM[3,1]-RM[4,1])), col=c("turquoise","red")) +
|
||
|
|
# theme_bw()
|
||
|
|
# print(p2)
|
||
|
|
}
|
||
|
|
CIsM <- t(matrix(POT, nrow=5))
|
||
|
|
CIsM <- cbind(CIsM, rep(x,nrow(CIsM)))
|
||
|
|
print(paste( testC, refC,sd(CIsM[,1])))
|
||
|
|
if (!exists("CIsM_all")) CIsM_all <- CIsM else CIsM_all <- rbind(CIsM_all, CIsM)
|
||
|
|
|
||
|
|
DevTargC3DF <- as.data.frame(t(matrix(DevTargC3, nrow=2)))
|
||
|
|
if (!exists("DevTargC3_")) DevTargC3_ <- DevTargC3DF else DevTargC3_ <- rbind(DevTargC3_, DevTargC3DF )
|
||
|
|
|
||
|
|
hhCIDF <- as.data.frame(t(matrix(hhCIV, nrow=6)))
|
||
|
|
if (!exists("hhCIDF_")) hhCIDF_ <- hhCIDF else hhCIDF_ <- rbind(hhCIDF_, hhCIDF )
|
||
|
|
}
|
||
|
|
|
||
|
|
CIsM_allDF <- as.data.frame(CIsM_all)
|
||
|
|
colnames(CIsM_allDF) <- c("potency","lCI","uCI","lpureCI","upureCI", "scheme")
|
||
|
|
CIsM_allDF$widthCI <- CIsM_allDF$uCI-CIsM_allDF$lCI
|
||
|
|
|
||
|
|
CIsM_allDF_ <- cbind(CIsM_allDF, DevTargC3_)
|
||
|
|
cn <- colnames(CIsM_allDF_)
|
||
|
|
cn[8:9] <- c("C3_REF","C3_TEST")
|
||
|
|
colnames(CIsM_allDF_) <- cn
|
||
|
|
CIsM_allDF_$C3_DIFF <- CIsM_allDF_$C3_TEST-CIsM_allDF_$C3_REF
|
||
|
|
|
||
|
|
cor(CIsM_allDF_$potency, CIsM_allDF_$C3_DIFF)
|
||
|
|
#cor(abs(log(CIsM_allDF_$potency)), (CIsM_allDF_[,9]-CIsM_allDF_[,8]))
|
||
|
|
plot( CIsM_allDF_$C3_DIFF,CIsM_allDF_$potency)
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
for (n in 0:2) {
|
||
|
|
te <- CIsM_allDF_[(n*N+1):((n+1)*N),]
|
||
|
|
print(paste("normal CI coverage", sum(log(te$lCI)<0 & log(te$uCI)>0) /N*100))
|
||
|
|
print(paste("pure CI coverage", sum(log(te$lpureCI)<0 & log(te$upureCI)>0) /N*100))
|
||
|
|
print(cor(te$potency, (te$C3_DIFF)))
|
||
|
|
te$logpot <- log(te$potency)
|
||
|
|
# plot(te$C3_DIFF,te$potency, pch=19, xlab="Difference at C3")
|
||
|
|
# abline(te$potency ~ te$C3_DIFF, col = "blue")
|
||
|
|
# lines(lowess(te$potency ~ te$C3_DIFF), col = "green")
|
||
|
|
scatterplot(logpot ~ C3_DIFF, data=te)
|
||
|
|
}
|
||
|
|
|
||
|
|
plot(te$C3_DIFF,te$logpot, pch=19, xlab="Difference at C3")
|
||
|
|
abline(te$logpot ~ te$C3_DIFF, col = "blue")
|
||
|
|
lines(lowess(te$logpot ~ te$C3_DIFF), col = "green")
|
||
|
|
scatterplot(logpot ~ C3_DIFF, data=te)
|
||
|
|
|
||
|
|
colnames(hhCIDF_) <- c("homomean", "lhomoCI", "uhomoCI","heteromean","lheteroCI","uheteroCI")
|
||
|
|
te <- cbind(CIsM_allDF, hhCIDF_)
|
||
|
|
|
||
|
|
|
||
|
|
p_histCI <- ggplot(CIsM_allDF, aes(x=widthCI, y=as.factor(scheme), fill=as.factor(scheme))) +
|
||
|
|
geom_density_ridges() +
|
||
|
|
theme_ridges()
|
||
|
|
p_histCI
|
||
|
|
|
||
|
|
p_histCI2 <- ggplot(CIsM_allDF, aes(x=widthCI, y=as.factor(scheme), fill=as.factor(scheme))) +
|
||
|
|
geom_density_ridges() +
|
||
|
|
theme_ridges()
|
||
|
|
p_histCI2
|
||
|
|
|
||
|
|
|
||
|
|
sum(Flags)/N*100
|
||
|
|
mean(SDs1)
|
||
|
|
mean(SDs2)
|
||
|
|
|
||
|
|
ro_new <- Calc_DilRes(sd_fac=0.1, log_concREF=log_concREF,
|
||
|
|
log_concTEST=log_concTEST,THEOconc = log(Conc)[4:11])
|
||
|
|
|
||
|
|
|