Machine Learning for Predicting Hyperglycemic Cases Induced by PD-1/PD-L1 Inhibitors

Algorithm 1

Key model setup algorithm.

	library(cvTools)
	library(foreach)
	library(doParallel)
	myfun = function (l) {
	library(e1071)
	#svm
	svm.fit = svm(reac∼., data = train, nu = l, cost = i, gamma = j, kernel = “radial,” type = “nu-regression”)
	svm.pre = ifelse(predict(svm.fit,train[, −9])>mean(predict(svm.fit,train[, −9])), 1, 0)
	reac = as.numeric(train[,“reac”])
	n = ifelse(svm.pre = = train[, 9], 1, 0)
	#result
	t = table(pred = svm.pre, true = train[, 9])
	A = as.vector(t)
	ac = (A[1] + A[4])/sum(A[1 : 4]) #Accuracy
	s = A[4]/(A[3] + A[4]) #Sensitive
	p = A[4]/(A[2] + A[4]) #precision
	f = 2/(1/p + 1/s) #F1 score
	pe = ((A[4] + A[2]) ^∗ (A[4] + A[3]) + (A[3] + A[1]) ^∗ (A[2] + A[1]))/(sum(A[1 : 4])^^2)
	k = (ac − pe)/(1 − pe) #Kappa
	#output
	c(l, ac, s, f, k, sum(n))
	}
	set.seed(80)
	R = 3
	K = 10
	cost.r = 3
	gamma.r = 10
	nu.r = seq(0.01, 1, 0.01)
	cv = cvFolds(NROW(d.train), K = K, R = R)
	af = foreach(r = 1: R,.combine = rbind) %do% {
	library(foreach)
	bf = foreach(k = 1: K,.combine = rbind) %do% {
	validation.idx = cv$subsets[which(cv$which = = k),r]
	train = d.train[-validation.idx,]
	validation = d.train[validation.idx,]
	cf = foreach (i = 1:cost.r,.combine = rbind) %do% {
	df = foreach (j = 1:gamma.r,.combine = rbind) %do% {
	library(snowfall)
	library(parallel)
	sfInit(parallel = T, cpus = detectCores() − 1)
	sfExport(“train,” “myfun,” “j,” “i,” “K,” “R”)
	a = sfLapply(nu.r, myfun)
	b = matrix(unlist(a), ncol = 6, byrow = T)
	sfStop()
	cbind(rep(j, nrow(b)), b)
	}
	cbind(rep(i,nrow(df)), df)
	} } }