7  The combined model

Here we present the predictive performance of the 2- to 10-cfRNA models trained in the 28wkGA samples from the combined dataset from discovery and validation cohort – we called these models as “combined models” or preterm+term models because discovery cohort is from the preterm delivery and the validation cohort from the term delivery.

Then, we compare their predictive performances to those from the 2- to 10-cfRNA models trained in the 28wkGA samples of discovery cohort – we call these models as the original models (or preterm models) as mentioned in the main text of the paper (see Figure 8.8). The 2- to 10-cfRNAs in the combined and the original models were chosen by Elastic net which was chosen as the best feature selection method from the 5-fold cross-validation, which is explained in Section 3.4 and Figure 8.7 A.

Firstly, we need to combine the 28wkGA dataset from both discovery and validation cohort. You may need to refer to Listing 3.1.

# combine the 28wkGA from discovery (preterm) and validation (term) dataset
x<-rbind(li.mat[["train"]][["28wk"]],li.mat[["test"]][["28wk"]])
mat.tr<-x # NB, the training dataset is now the combined dataset

li.num<-2:10 %>% as.list
names(li.num)=paste0("F",2:10)

# Featue selection via ElasticNet only 
dl.combined.models<-
  lapply(li.num, function(my.num){
    message(paste("Num:",my.num,"Method: ENET"))
    dt.enet<-get_enet_coef(x=x, "combined", my.num=my.num)
    dt.enet[method=="ENet2"]
  })
save(dl.combined.models, file="RData/dl.combined.models.core17.RData")

## Combined models validated on POPS and & Munchel
dl.combined.result<-lapply(dl.combined.models, function(dt.model){
  dt.final.model<-dt.model[order(feature)][,.(.N,features=paste(feature,collapse=",")),method][order(features)]
  dt.final<-dt.final.model[,.(.N,methods=paste(method,collapse=',')),features]

  #######################################################################
  # get LPOCV/AUC from the preterm dataset (NB, 28wk-preterm: training) #
  #######################################################################
  dt.final.result<-lapply(dt.final$methods, function(my.methods){
                        ############################################
                        # fit the model using the training dataset #
                        ############################################
                        my.feature<-dt.final[methods==my.methods]$features %>% strsplit(",") %>% unlist
                        df.mat.tr<-mat.tr[,c(my.feature,'y')] %>% as.data.frame  # training set
                        my.model<-glm(y~. , data = df.mat.tr, family = "binomial")

                        ## preterm+term (NB, 28wk: dataset where the model was built)
                        dt.foo0<-lapply(c("12wk","20wk","28wk"), function(my.GA){
                          message(paste("preterm+term",my.methods,my.GA,sep=":"))
                          x<-rbind(li.mat[["train"]][[my.GA]], li.mat[["test"]][[my.GA]])
                          my.fold<-paste0(my.GA,"(preterm+term)")
                          cbind(`methods`=my.methods,
                                get_cv_glm2(x=x,my.fold=my.fold,my.model=my.model,my.feature=my.feature)
                          )
                        }) %>% rbindlist

                        ## preterm (NB, 28wk: dataset where the model was built)
                        dt.foo1<-lapply(c("12wk","20wk","28wk"), function(my.GA){
                          message(paste("preterm",my.methods,my.GA,sep=":"))
                          x<-li.mat[["train"]][[my.GA]]
                          my.fold<-paste0(my.GA,"(preterm)")
                          cbind(`methods`=my.methods,
                                get_cv_glm2(x=x,my.fold=my.fold,my.model=my.model,my.feature=my.feature)
                          )
                        }) %>% rbindlist

                        ## term (validation)
                        dt.foo2<-lapply(c("12wk","20wk","28wk","36wk"), function(my.GA){
                          message(paste("term",my.methods,my.GA,sep=":"))
                          x<-li.mat[["test"]][[my.GA]]
                          my.fold<-paste0(my.GA,"(term)")
                          cbind(`methods`=my.methods,
                                get_cv_glm2(x=x,my.fold=my.fold,my.model=my.model,my.feature=my.feature)
                          )
                        }) %>% rbindlist
                        
                        ## Munchel 
                        message(paste("Munchel",my.methods,sep=":"))
                        x<-li.mat[["munchel"]]
                        my.fold<-"Munchel"
                        dt.foo3<-cbind(`methods`=my.methods,
                                get_cv_glm2(x=x,my.fold=my.fold,my.model=my.model,my.feature=my.feature)
                        )

                        rbind(dt.foo0, dt.foo1,dt.foo2,dt.foo3)
                    }) %>% rbindlist
  dt.final.result<-dt.final.result[order(fold,-AUC_test)]
  dt.final.result
}) # end of dl.enet.models
save(dl.combined.result, file="RData/dl.combined.result.core17.RData")