forked from QSong-github/scGCN
-
Notifications
You must be signed in to change notification settings - Fork 1
/
data_preprocess_utility.R
278 lines (255 loc) · 13.3 KB
/
data_preprocess_utility.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
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
#' This functions provides the optional way to generate scGCN labels
#' @param Dat1 reference data; rows are genes and columns are cells
#' @param Dat2 query data; rows are genes and columns are cells
#' @param Lab1 label of reference data; rows are genes and columns are cells
#' @param Lab2 query data; rows are genes and columns are cells
#' @param inter_graph inter-dataset graph in input folder
#' @return This function returns a list of two statistical scores: entropy score and enrichment score that unknown cells have higher entropy score and lower enrichment score
#' @examples: first load count.list and label.list from folder "example_data" and run 'data_preprocess.R'
#' then run 'metrics(dat1,lab1,dat2,lab2,inter_graph,cluster)'
suppressMessages(library(Seurat)); suppressMessages(library(entropy))
metrics <- function(lab1,inter_graph,clusters){
inter.graph <- inter_graph; hc <- clusters
#' ------------ check cluster enrichment in cell types ------------
#' Examine in clusters
#' check enrichment
eval1 <- sapply(names(table(hc)),function(name){
ps1 <- which(hc==name)
foo <- table(lab1[inter_graph[which(inter_graph[,2]%in%ps1),1],1])
tem <- table(lab1)/sum(table(lab1))
if (length(foo)>0){
temp <- unlist(sapply(1:length(foo),function(i){
ind <- grep(names(foo)[i],names(tem))
r1 <- foo[i]/sum(foo)
return (r1/tem[ind])}))
return (max(temp)/sum(temp))
} else {return (0)}
})
#' check mixtureness
eval2 <- sapply(names(table(hc)),function(name){
ps1 <- which(hc==name)
foo <- table(lab1[inter_graph[which(inter_graph[,2]%in%ps1),1],1])
tem <- table(lab1)/sum(table(lab1))
if (length(foo)>0){
temp <- unlist(sapply(1:length(foo),function(i){
ind <- grep(names(foo)[i],names(tem))
r1 <- foo[i]/sum(foo)
return (r1/tem[ind])}))
temp2 <- entropy.empirical(temp,unit='log')
return (temp2)
} else {return (0) }
})
return (list(eval1=eval1,eval2=eval2))
}
#' This functions provides the optional way to generate scGCN labels
#' @param Dat1 reference data; rows are genes and columns are cells
#' @param Dat2 query data; rows are genes and columns are cells
#' @param Lab1 label of reference data; rows are genes and columns are cells
#' @param Lab2 query data; rows are genes and columns are cells
#' @return This function returns files saved in folders "input" & "process_data"
#' @export: all files are saved in current path
#' @examples: load count.list and label.list from folder "example_data"
#' save_processed_data(count.list,label.list)
GenerateGraph <- function(Dat1,Dat2,Lab1,K,check.unknown){
object1 <- CreateSeuratObject(counts=Dat1,project = "1",assay = "Data1",
min.cells = 0,min.features = 0,
names.field = 1,names.delim = "_")
object2 <- CreateSeuratObject(counts=Dat2,project = "2",assay = "Data2",
min.cells = 0,min.features =0,names.field = 1,
names.delim = "_")
objects <- list(object1,object2)
objects1 <- lapply(objects,function(obj){
obj <- NormalizeData(obj,verbose=F)
obj <- FindVariableFeatures(obj,
selection.method = "vst",
nfeatures = 2000,verbose=F)
obj <- ScaleData(obj,features=rownames(obj),verbose=FALSE)
obj <- RunPCA(obj, features=rownames(obj), verbose = FALSE)
return(obj)})
#' Inter-data graph
object.nn <- FindIntegrationAnchors(object.list = objects1,k.anchor=K,verbose=F)
arc=object.nn@anchors
d1.arc1=cbind(arc[arc[,4]==1,1],arc[arc[,4]==1,2],arc[arc[,4]==1,3])
grp1=d1.arc1[d1.arc1[,3]>0,1:2]-1
if (check.unknown){
obj <- objects1[[2]]
obj <- RunPCA(obj, features = VariableFeatures(object = obj),npcs=30,verbose=F)
obj <- FindNeighbors(obj,verbose=F)
obj <- FindClusters(obj, resolution = 0.5,verbose=F)
hc <- Idents(obj); inter.graph=grp1+1
scores <- metrics(lab1=Lab1,inter_graph=inter.graph,clusters=hc)
saveRDS(scores,file='./input/statistical_scores.RDS')
}
#' Intra-data graph
d2.list <- list(objects1[[2]],objects1[[2]])
d2.nn <- FindIntegrationAnchors(object.list =d2.list,k.anchor=K,verbose=F)
d2.arc=d2.nn@anchors
d2.arc1=cbind(d2.arc[d2.arc[,4]==1,1],d2.arc[d2.arc[,4]==1,2],d2.arc[d2.arc[,4]==1,3])
d2.grp=d2.arc1[d2.arc1[,3]>0,1:2]-1
final <- list(inteG=grp1,intraG=d2.grp)
return (final)
}
#' This function normalize count data matrix from refrence and query set
#' @param count.list list of (1) reference data and (2) query data, rows are genes and columns are cells
#' @return This function returns normalized data list
#' @export
#' @examples
#' normalize_data(count.list)
normalize_data <- function(count.list){
norm.list <- vector('list')
var.features <- vector('list')
for ( i in 1:length(count.list)){
norm.list[[i]] <- as.matrix(Seurat:::NormalizeData.default(count.list[[i]],verbose=F))
#' select variable features
hvf.info <- Seurat:::FindVariableFeatures.default(count.list[[i]],selection.method='vst',verbose=F)
hvf.info <- hvf.info[which(x = hvf.info[, 1, drop = TRUE] != 0), ]
hvf.info <- hvf.info[order(hvf.info$vst.variance.standardized, decreasing = TRUE), , drop = FALSE]
var.features[[i]] <- head(rownames(hvf.info), n = 2000)
}
#' select variable features
sel.features <- selectHVFeature(count.list,var.features)
return (list(norm.list,sel.features))}
#' This function returns scaled data matrix from refrence and query set
#' @param count.list list of raw counts of (1) reference data and (2) query data, rows are genes and columns are cells
#' @param norm.list list of normalized (1) reference data and (2) query data, rows are genes and columns are cells
#' @param hvg.features variable features of data list
#' @return This function returns scaled data list
#' @export
#' @examples
#' scale_data(count.list,norm.list,hvg.features)
scale_data <- function(count.list,norm.list,hvg.features){
scale.list <- lapply(norm.list,function(mat){
Seurat:::ScaleData.default(object = mat, features = hvg.features,verbose=F)})
scale.list <- lapply(1:length(count.list),function(i){
return (scale.list[[i]][na.omit(match(rownames(count.list[[i]]),rownames(scale.list[[i]]))),])})
return (scale.list)}
#' This function returns a set of highly variable features
#' @param count.list list of raw counts of (1) reference data and (2) query data, rows are genes and columns are cells
#' @param var.features variable features of data list
#' @return This function returns a common set of highly variable features
#' @export
#' @examples
#' selectHVFeature(count.list,var.features)
selectHVFeature <- function(count.list,var.features,nfeatures = 2000){
var.features1 <- unname(unlist(var.features))
var.features2 <- sort(table(var.features1), decreasing = TRUE)
for (i in 1:length(count.list)) {
var.features3 <- var.features2[names(var.features2) %in% rownames(count.list[[i]])]}
tie.val <- var.features3[min(nfeatures, length(var.features3))]
features <- names(var.features3[which(var.features3 > tie.val)])
if (length(features) > 0) {
feature.ranks <- sapply(features, function(x) {
ranks <- sapply(var.features, function(y) {
if (x %in% y) {
return(which(x == y))
}
return(NULL)
})
median(unlist(ranks))
})
features <- names(sort(feature.ranks))
}
features.tie <- var.features3[which(var.features3 == tie.val)]
tie.ranks <- sapply(names(features.tie), function(x) {
ranks <- sapply(var.features, function(y) {
if (x %in% y) {return(which(x == y))}
return(NULL)
})
median(unlist(ranks))
})
features <- c(features, names(head(sort(tie.ranks), nfeatures - length(features))))
return(features)
}
#' This functions takes refrence data and labels to identify variable gene features
#' @param data reference data; rows are genes and columns are cells
#' @param label data frame with rownames identical with colnames of data; the first column is cell type
#' @param nf number of variable features
#' @return This function returns variable gene features using ANOVA
#' @export
#' @examples
#' select_feature(data=reference.data,label=reference.label)
select_feature <- function(data,label,nf=2000){
M <- nrow(data); new.label <- label[,1]
pv1 <- sapply(1:M, function(i){
mydataframe <- data.frame(y=as.numeric(data[i,]), ig=new.label)
fit <- aov(y ~ ig, data=mydataframe)
summary(fit)[[1]][["Pr(>F)"]][1]})
names(pv1) <- rownames(data)
pv1.sig <- names(pv1)[order(pv1)[1:nf]]
egen <- unique(pv1.sig)
return (egen)
}
#' This functions takes refrence data and labels to identify variable gene features
#' @param count_list list of reference data and query data; rows are genes and columns are cells
#' @param label_list list of reference label and query label (if any), both are data frames with rownames identical with colnames of data; the first column is cell type
#' @return This function returns four elements, including the normalized data, scaled data, hvg features, and selected features
#' @export
#' @examples
#' pre_process(count_list,label_list)
pre_process <- function(count_list,label_list){
sel.features <- select_feature(count_list[[1]],label_list[[1]])
count_list_new <- list(count_list[[1]][sel.features,],count_list[[2]][sel.features,])
return (count_list_new)
}
#' This functions takes raw counts and labels of reference/query set to generate scGCN training input
#' @param count.list list of reference data and query data; rows are genes and columns are cells
#' @param label.list list of reference label and query label (if any), both are data frames with rownames identical with colnames of data; the first column is cell type
#' @return This function returns files saved in folders "input" & "process_data"
#' @export: all files are saved in current path
#' @examples: load count.list and label.list from folder "example_data"
#' save_processed_data(count.list,label.list)
save_processed_data <- function(count.list,label.list,Rgraph=TRUE,check_unknown=FALSE){
count.list <- pre_process(count_list=count.list,label_list=label.list)
#' save counts data to certain path: 'input'
dir.create('input');
write.csv(t(count.list[[1]]),file='input/Data1.csv',quote=F,row.names=T)
write.csv(t(count.list[[2]]),file='input/Data2.csv',quote=F,row.names=T)
#' optional graph: R genreated graph has minor differnce with python, user can choose the one with better performance
if (Rgraph){
#' use R generated graph
new.dat1 <- count.list[[1]]; new.dat2 <- count.list[[2]]
new.lab1 <- label.list[[1]]; new.lab2 <- label.list[[2]]
graphs <- suppressWarnings(GenerateGraph(Dat1=new.dat1,Dat2=new.dat2,
Lab1=new.lab1,K=5,
check.unknown=check_unknown))
write.csv(graphs[[1]],file='input/inter_graph.csv',quote=F,row.names=T)
write.csv(graphs[[2]],file='input/intra_graph.csv',quote=F,row.names=T)
write.csv(new.lab1,file='input/label1.csv',quote=F,row.names=F)
write.csv(new.lab2,file='input/label2.csv',quote=F,row.names=F)
} else {
#' use python generated graph
dir.create('results')
#' @param norm.list normalized data
res1 <- normalize_data(count.list)
norm.list <- res1[[1]]; hvg.features <- res1[[2]];
#' @param scale.list scaled data
scale.list <- scale_data(count.list,norm.list,hvg.features)
outputdir <- 'process_data'; dir.create(outputdir)
write.csv(hvg.features,file=paste0(outputdir,'/sel_features.csv'),quote=F,row.names=F)
N <- length(count.list)
for (i in 1:N){
df = count.list[[i]]
if (!dir.exists(paste0(outputdir,'/count_data'))){dir.create(paste0(outputdir,'/count_data'))}
file.name=paste0(outputdir,'/count_data/count_data_',i,'.csv')
write.csv(df,file=file.name,quote=F)
}
for (i in 1:N){
df = label.list[[i]]
if (!dir.exists(paste0(outputdir,'/label_data'))){dir.create(paste0(outputdir,'/label_data'))}
file.name=paste0(outputdir,'/label_data/label_data_',i,'.csv')
write.csv(df,file=file.name,quote=F)
}
for (i in 1:N){
df = norm.list[[i]]
if (!dir.exists(paste0(outputdir,'/norm_data'))){dir.create(paste0(outputdir,'/norm_data'))}
file.name=paste0(outputdir,'/norm_data/norm_data_',i,'.csv')
write.csv(df,file=file.name,quote=F)
}
for (i in 1:N){
df = scale.list[[i]]
if (!dir.exists(paste0(outputdir,'/scale_data'))){dir.create(paste0(outputdir,'/scale_data'))}
file.name=paste0(outputdir,'/scale_data/scale_data_',i,'.csv')
write.csv(df,file=file.name,quote=F)
}
}
}