-
Notifications
You must be signed in to change notification settings - Fork 0
/
ModelPredict.R
176 lines (160 loc) · 6.96 KB
/
ModelPredict.R
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
# This is the script for predicting and ranking
# Author: Shuyang Zhao (shuyang.zhao@cchmc.org)
source("https://raw.githubusercontent.com/xu-lab/SINCERA/umi/R/functions.R")
library(plyr)
library(dplyr)
library(Matrix)
library(reshape2)
library(openxlsx)
library(glmnet)
library(caret)
library(ROCR)
############################ Loading prepared data package ######################################
celltype <- "AT1"
#load pre-stored traiing and testing set
load(file=paste("GS.UMI/",celltype,"/",celltype,".model.data.package.Rda",sep=""))
#variable transformation
trainset$peak <- as.numeric(as.character(trainset$peak))
for(i in 1:length(list.test)){
list.test[[i]]$peak <- as.numeric(as.character(list.test[[i]]$peak))
}
dim(trainset)
class(trainset$peak)
##############################################################################################################
################################################ model fiting ###############################################
##############################################################################################################
trainset <- trainset[,-1]
#predictor variables
x <- model.matrix(y~.,trainset)[,-1]
#response variable
y <- trainset$y
#############ELASTIC NET################
#set.seed(12345)
#"caret" train
elastic <- train(
y ~., data = trainset, method = "glmnet",
trControl = trainControl("cv", number = 10),
tuneLength = 10
)
#model coefficients
elastic$bestTune$alpha
elastic$bestTune$lambda
#fitting the model
coef(elastic$finalModel, elastic$bestTune$lambda)
#using alpha, lambda value from "caret" and fit glmnet() model
elnet.model <- glmnet(x,y,family="binomial",
alpha=elastic$bestTune$alpha,
lambda =elastic$bestTune$lambda)
############################################################################################
#Looping throught testing sets
list.predict <- list()
for(z in 1:length(list.test)){
testset <- list.test[[z]]
#get test data
testset <- testset[,-1]
testset$y <- rep(0,nrow(testset))
x_test <- model.matrix(y~.,testset)[,-1]
#dim(x_test)
#predict probability
elnet_prob <- predict(elnet.model,newx = x_test,type="response")
elnet_prob <- as.data.frame(elnet_prob)
genes <- unlist(lapply(rownames(elnet_prob), extract_field, field=3, delim="."))
elnet_prob$gene <- genes
#loop through all alphas and lambdas parameters
grid<-expand.grid(.alpha=seq(0,1,by=.1),.lambda=seq(0,1,by=.1))
for (i in unique(grid$.alpha)){
for (j in unique(grid$.lambda)){
model <- glmnet(x,y,family="binomial",
alpha=i,
lambda =j)
prob <- predict(model,newx = x_test,type="response")
prob <- as.data.frame(prob)
col.name <- paste("alpha",i, "lambda",j, sep=".")
elnet_prob[,col.name] <- prob$s0
}
}
colnames(elnet_prob)[1] <- "alpha.best.lambda.best"
#save(elnet_prob,file=paste(celltype,"elnet_prob.rda",sep="."))
#ROC to test parameters performance
ROC.genes <- rownames(genes.train)[which(genes.train$group == "positive")]
index <- which(elnet_prob$gene %in% ROC.genes)
length(index)
elnet_prob.1 <- elnet_prob[,-2]
#rownames(elnet_prob.1)[index]
auc.perf <- data.frame(colnames(elnet_prob.1),auc = 0)
colnames(auc.perf) <- c("parameters","auc")
for (i in 1:ncol(elnet_prob.1)){
i.cells <- data.frame(p = elnet_prob.1[,i], s=0)
#get marker genes
i.cells[index,]$s <- 1
#AUC
i.pred <- prediction(i.cells$p, i.cells$s)
i.perf <- performance(i.pred,"auc")
i.auc <- i.perf@y.values[[1]]
auc.perf[i,]$auc <- i.auc
}
#locate the parameters with the best result
param.select <- as.character(auc.perf[which(auc.perf$auc == max(auc.perf$auc)),]$parameters)
###################################################################################################################
##################################################### RANKING #####################################################
###################################################################################################################
#print(param.select)
if(length(param.select) >1){
print(paste("Converting multiple parameters for",names(list.test)[z],sep=" "))
predict.best <- elnet_prob[,c("gene",param.select)]
tmp.rank <- matrix(0,nrow=nrow(predict.best),ncol=length(param.select))
rownames(tmp.rank) <- genes
tmp.rank <- as.data.frame(tmp.rank)
colnames(tmp.rank) <- param.select
for(j in 2:ncol(predict.best)){
predict.best <- predict.best[order(predict.best[,j],decreasing = T),]
rank <- as.data.frame(as.matrix(cbind(predict.best$gene,c(1:nrow(predict.best)))))
rownames(rank) <- rank$V1
rank <- rank[rownames(tmp.rank),]
tmp.rank[,j-1] <- as.numeric(as.character(rank$V2))
}
tmp.rank$avg.rank <- rowSums(tmp.rank)/length(param.select)
tmp.rank <- tmp.rank[order(tmp.rank$avg.rank,decreasing = F),]
tmp.rank$rank <- c(1:nrow(tmp.rank))
tmp.rank$gene <- rownames(tmp.rank)
tmp.rank$source <- rep(unlist(lapply(names(list.test)[z], extract_field, field=1, delim=".")),nrow(tmp.rank))
tmp.rank$tps <- rep(unlist(lapply(names(list.test)[z], extract_field, field=2, delim=".")),nrow(tmp.rank))
tmp.rank <- tmp.rank[,-which(colnames(tmp.rank)%in%param.select)]
tmp.rank <- tmp.rank[,c(3,4,5,1,2)]
rownames(tmp.rank) <- paste(tmp.rank$source,tmp.rank$tps,tmp.rank$gene)
list.predict[[z]] <- tmp.rank
names(list.predict)[z] <- names(list.test)[z]
write.table(tmp.rank,paste(celltype,names(list.test)[z],"predict.best.txt",sep="."), sep="\t",quote = F)
}
else{
print(paste("The best parameters for",names(list.test)[z],"is",param.select,sep=" "))
predict.best <- elnet_prob[,c("gene",param.select)]
predict.best$source <- unlist(lapply(rownames(predict.best), extract_field, field=1, delim="."))
predict.best$tps <- unlist(lapply(rownames(predict.best), extract_field, field=2, delim="."))
predict.best <- predict.best[,c(1,3,4,2)]
colnames(predict.best) <- c("gene","source","tps","prob")
predict.best <- predict.best[order(predict.best$prob,decreasing = T),]
predict.best$rank <- c(1:nrow(predict.best))
list.predict[[z]] <- predict.best
names(list.predict)[z] <- names(list.test)[z]
#output individual sample ranking
#write.table(predict.best,paste(celltype,names(list.test)[z],"predict.best.txt",sep="."), sep="\t",quote = F)
}
}
#calculate the average rank
ranks <- matrix(0,nrow=length(genes),ncol=length(list.predict))
rownames(ranks) <- genes
ranks <- as.data.frame(ranks)
for(i in 1:length(list.predict)){
tmp <- list.predict[[i]]
rownames(tmp) <- as.character(tmp$gene)
a <- tmp[rownames(ranks),which(colnames(tmp)=="rank")]
ranks[,i] <- a
colnames(ranks)[i] <- names(list.predict)[i]
}
ranks$sum.rank <- rowSums(ranks)
ranks$avg.rank <- ranks$sum.rank/length(list.test)
#order based on average ranks
ranks <- ranks[order(ranks$avg.rank,decreasing = F),]
#output final ranking
write.table(ranks,paste(celltype,"ranks.txt",sep="."), sep="\t",quote = F)