require(parallel)
require(lme4) 

load("mdata.RData") #file including DNA methylation beta-values

#optional: only use a subset of the CpGs included in mdata file
subset <- read.csv("CpGs.csv", header=T, strings=F) 
head(subset)
mdata <- mdata[,colnames(mdata) %in% subset$CpGs]


pdata<-read.csv("pdata.csv", header=T, strings=F) #file including participant data

mdata <- mdata[rownames(mdata) %in% pdata$id,]  
pdata <- pdata[pdata$id %in% rownames(mdata),]
pdata <- pdata[match(rownames(mdata),pdata$id),]  

##Exposure= Cardio-metabolic trait of interest
#model 1
fm_char <- "y ~ Exposure + sex + Age + Granulocytes + Lymphocytes + Monocytes (1|Array_Number) + (1|Position_on_Array)"

#model 2 including BMI + relevant medication
fm_char <- "y ~ Exposure + sex + Age + BMI + medication + Granulocytes + Lymphocytes + Monocytes (1|Array_Number) + (1|Position_on_Array)" 

fm <- as.formula(fm_char)

fm_fixed <- gsub("\\(1\\|", "", fm_char)         
fm_fixed <- as.formula(gsub("\\)", "", fm_fixed))       

method = "LMM"
if (method == "LMM") {
  pdata[,"y"] <- rnorm(nrow(pdata))
  DFE <- lm(fm_fixed, data=pdata)$df.residual
  
  # LMM
  doOne <- function(i) {
    pdata$y <- as.numeric(mdata[,i])
    
    betamean<-mean(pdata$y,na.rm=T)
    betasd<-sd(pdata$y,na.rm=T)      
    result <- c(NA, NA, NA, length(na.omit(pdata$y)), betamean, betasd)
    buf <- tryCatch({
      mresult <- lmer(fm, data=pdata)
      reduced_y <- mresult@frame[, "y"]
      tbl <- summary(mresult)
      # Automatic detection between new and old lme4
      if (any(names(attributes(summary(tbl))) == "coefs")) {
        tbl <- tbl@coefs
      } else {
        tbl <- tbl$coef
      }
      # Returns Effect size, Standard Error, P-value, N
      result <- c(tbl[2,1], tbl[2,2], 2*pt(abs(tbl[2,3]), df=DFE, lower.tail=FALSE), length(reduced_y), betamean, betasd)          
    }, error = function(e) {
      cat("WARNING: LMM failed for probe",i,"\n")
      print(e)
    }, finally={})
    
    return (result)
  }
} 

result_all <- do.call(rbind, mclapply(1:ncol(mdata), doOne));
result_all <- data.frame(result_all)
result_all$ProbeID <- colnames(mdata)
result_all <- result_all[,c(ncol(result_all), 1:(ncol(result_all)-1))]
colnames(result_all) <- c("ProbeID", "Effect", "SE", "P", "N", "betaMean", "betaSD")

write.csv(result_all,"CpGs_exposure.csv",quote=F,row.names=F)