-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathBrowseData_Dom1to3.R
269 lines (223 loc) · 11.8 KB
/
BrowseData_Dom1to3.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
rm(list=ls())
source('D:/GitHub/R-General/.Rprofile')
source('D:/GitHub/R-Informatics-private/HuMoments.R')
source('D:/GitHub/R-Informatics-private/Zernike.R')
source('D:/GitHub/R-Cytoprofiling/PreProcessingHelpers.R')
# source('~/.Rprofile')
# source('~/Public/DropBox/GitHub/R-Informatics-private/HuMoments.R')
# source('/Users/jaywarrick/Public/DropBox/GitHub/R-Informatics-private/Zernike.R')
# source('~/Public/DropBox/GitHub/R-Cytoprofiling/PreProcessingHelpers.R')
#sourceGitHubFile(user='jaywarrick', repo='R-Cytoprofiling', branch='master', file='PreProcessingHelpers.R')
library(data.table)
library(foreign)
# A1,2,3 - WT
# A4,5,6 - NES
# A7,9,10 - WT SecOnly
# A8,11,12 - NES SecOnly
#
# B10,11,12 - WT
# B7,8,9 - NES
# B3,4,6 - WT SecOnly
# B1,2,5 - NES SecOnly
#dir1 <- '/Volumes/Seagate Backup Plus Drive/Data Tables/Dominique 1'
dir1 <- 'G:/Data Tables/Dominique 1'
MTFileList1 <- c('x0_y0.csv')
WTFileList1 <- c('x1_y0.csv')
#dir2 <- '/Volumes/Seagate Backup Plus Drive/Data Tables/Dominique 2'
dir2 <- 'G:/Data Tables/Dominique 2'
MTFileList2 <- c('x0_y0.csv','x0_y1.csv','x0_y2.csv','x0_y3.csv')
WTFileList2 <- c('x0_y4.csv','x0_y5.csv','x0_y6.csv')
#dir3 <- '/Volumes/Seagate Backup Plus Drive/Data Tables/Dominique 3'
dir3 <- 'G:/Data Tables/Dominique 3'
MTFileList3 <- c('x0_y3.csv','x0_y4.csv','x0_y5.csv')#,'x1_y6.csv','x1_y7.csv','x1_y8.csv')
WTFileList3 <- c('x0_y0.csv','x0_y1.csv','x0_y2.csv')#,'x1_y9.csv','x1_y10.csv','x1_y11.csv')
##### DATA PREPROCESSING #####
set.seed(1234)
n <- 3000
tableList <- list()
tableList <- append(tableList, getTableList(dir1, MTFileList1, 'MT', expt=1, sampleSize = n))
tableList <- append(tableList, getTableList(dir1, WTFileList1, 'WT', expt=1, sampleSize = n))
tableList <- append(tableList, getTableList(dir2, MTFileList2, 'MT', expt=2, sampleSize = n))
tableList <- append(tableList, getTableList(dir2, WTFileList2, 'WT', expt=2, sampleSize = n))
tableList <- append(tableList, getTableList(dir3, MTFileList3, 'MT', expt=3, sampleSize = n))
tableList <- append(tableList, getTableList(dir3, WTFileList3, 'WT', expt=3, sampleSize = n))
x1 <- rbindlist(tableList, use.names=TRUE)
# Check number sampled
#length(unique(x1$cId[grepl('2.x',x1$cId,fixed=T) & x1$Class=='MT']))
fixColNames(x1)
# Make things easier to peruse
setorder(x1, Expt, file, Id, Label, MaskChannel, Measurement, ImageChannel)
x1$Id <- as.character(x1$Id) # Avoid standardizing the Id
save(x1, file='/Volumes/Seagate Backup Plus Drive/Data Tables/x1_sampled.Rdata')
save(x1, file='G:/Data Tables/x1_sampled.Rdata')
load(file='G:/Data Tables/x1_sampled.Rdata')
set.seed(1234)
x1b <- fixLongTableStringsInCol(x1, 'Measurement')
#x1b <- removeMeasurementNamesContaining(x1b,'DNZ')
#x1b <- removeMeasurementNamesContaining(x1b,'NUCw')
x1b <- removeMeasurementNamesContaining(x1b,'ZernikeCircleX')
x1b <- removeMeasurementNamesContaining(x1b,'ZernikeCircleY')
x1b <- removeMeasurementNamesContaining(x1b,'DNZernikeInnerCircleX')
x1b <- removeMeasurementNamesContaining(x1b,'DNZernikeOuterCircleY')
x1b <- replaceSubStringInAllRowsOfCol(x1b, '395 X 455M', '390 X 440', 'ImageChannel')
x1b <- replaceSubStringInAllRowsOfCol(x1b, '485 X 525 M', '485 X 525', 'ImageChannel')
x1b <- replaceSubStringInAllRowsOfCol(x1b, '560 X 607 M', '560 X 607', 'ImageChannel')
x1b <- replaceSubStringInAllRowsOfCol(x1b, '650 X 705 M', '648 X 684', 'ImageChannel')
x1b <- x1b[ImageChannel != '560 X 607']
#x1b <- x1b[MaskChannel == 'WholeCell']
# Get rid of cells that have some NA data
naData <- unique(x1b[!is.finite(Value)]$cId)
x1b <- x1b[!(cId %in% naData)]
goodData <- unique(x1b[Measurement == 'ZernikeMag11_NUCwFIXED']$cId)
x1b <- x1b[(cId %in% goodData)]
# Do some calculations
x2 <- intIntensityNormalizeCentralMoments(x1b)
x2 <- meanNormalizeZernikeMoments(x2)
x2 <- calculateHuMoments(x2)
x2 <- calculateZernikeDotProduct(x2)
# Tempororarily make the table wide to calculate averages of Haralick over the different directions
x2 <- getWideTable(x2)
x2 <- calculateRMSofHaralick(x2)
x2 <- removeExtraneousColumns(x2)
# Get our long table back and reorder
x3 <- getLongTableFromTemplate(x2, x1b)
setorder(x3, Expt, file, Id, Label, MaskChannel, Measurement, ImageChannel)
# Perform robust standardization (x-median)/mad (entertain idea of not applying to histogram bins)
x3 <- standardizeLongData(x3)
# Generate a table of differences between measures for each MaskChannel/ImageChannel/Measurement combination
x3 <- refactor(x3)
diffs <- calculateChannelDifferences(x3)
# Standardize the difference data
diffs <- standardizeLongData(diffs)
# Merge it with the original dataset, merging MaskChannel and ImageChannel into MeasurementName
x3 <- rbindlist(list(x3,diffs), use.names = TRUE)
x3$Measurement <- paste(x3$Measurement, x3$MaskChannel, x3$ImageChannel, sep='_')
x3[,MaskChannel:=NULL]
x3[,ImageChannel:=NULL]
# Get a wide table for machine learning and plotting
x3 <- getWideTable(x3)
# Fix naming issues introduced by merging MaskChannel and ImageChannel names with Measurement name
x3 <- fixColNames(x3)
# Perform final sorting of columns of data for easier perusing
x3 <- sortColsByName(x3)
# shinyData[,lapply(.SD, function(x){if(is.factor(x)){return(as.factor(x))}else{return(x)}})]
shinyData <- copy(x3)
getNumericCols(shinyData)[as.logical(as.vector(shinyData[,lapply(.SD, function(x){length(which(!is.finite(x)))>0}),.SDcols=getNumericCols(shinyData)]))]
# Write the data for potential analysis outside R
#write.csv(shinyData, file='/Users/jaywarrick/Documents/MMB/Grants/2016 - RO1 Cytoprofiling/shinyData_Dom3.csv', row.names=FALSE)
write.csv(shinyData, file='G:/Data Tables/shinyData_Dom1to3_r1234.csv', row.names=FALSE)
#shinyData <- read.csv(file='/Users/jaywarrick/Documents/MMB/Grants/2016 - RO1 Cytoprofiling/test.csv')
# Look at the data
#browseShinyData()
###### RANDOM FOREST MACHINE LEARNING #####
library(randomForest)
# Need to remove names like Id, Label, ImRow, ImCol, Z,
dataToTest <- shinyData[Expt==1]
dataToTest[, c('cId','Id','Label','ImCol','ImRow','Z','Loc','file','Expt'):=NULL]
removeColsWithInfiniteVals(dataToTest)
dataToTest$Class <- as.factor(dataToTest$Class)
dataToTrain <- shinyData[Expt==3]
dataToTrain[, c('cId','Id','Label','ImCol','ImRow','Z','Loc','file','Expt'):=NULL]
removeColsWithInfiniteVals(dataToTrain)
dataToTrain$Class <- as.factor(dataToTrain$Class)
#write.csv(dataToTest, file='/Users/jaywarrick/Documents/MMB/Grants/2016 - RO1 Cytoprofiling/dataToTest_Dom2.csv', row.names = FALSE)
write.csv(dataToTest, file='G:/Data Tables/dataToTest_Dom1to3_r1234.csv', row.names = FALSE)
dataToTest <- fread('G:/Data Tables/dataToTest_Dom1to3_r1234.csv')
# Set the random seed to reproduce results
set.seed(1234)
# Learn the trees
rf <- randomForest(formula= Class ~ ., data=dataToTrain, ntree=100, importance=TRUE, proximity=TRUE, do.trace=TRUE, keep.forest=TRUE)
rf2 <- randomForest(formula= Class ~ ., data=dataToTrain, ntree=25, maxnodes=5, importance=TRUE, proximity=TRUE, do.trace=TRUE, keep.forest=TRUE)
rf3 <- randomForest(formula= Class ~ ., data=dataToTrain, ntree=25, importance=TRUE, proximity=TRUE, do.trace=TRUE, keep.forest=TRUE)
# Creat interactive plot to browse importance results
length(which(predict(rf2, dataToTest)==dataToTest$Class))/nrow(dataToTest)
library(plotly)
rfImp <- data.frame(rf2$importance)
rfImp$name <- row.names(rfImp)
rfImp <- rfImp[order(rfImp$MeanDecreaseAccuracy, decreasing=TRUE),]
plot(rfImp$MeanDecreaseAccuracy, pch=20, col='deepskyblue3', ylab='Importance', xlab='Feature Rank')
plot_ly(rfImp, mode='markers', x=row.names(rfImp), y=rfImp$MeanDecreaseAccuracy, text=row.names(rfImp))
layout(hovermode="closest")
rfImp$RelMeanDecreaseAccuracy <- rfImp$MeanDecreaseAccuracy/(max(rfImp$MeanDecreaseAccuracy))
rfImp[1:50,c('name','RelMeanDecreaseAccuracy')]
which(grepl('Dot',rfImp$name))
rfImp$name[which(grepl('Dot',rfImp$name))]
i <- 1
#dir.create('/Users/jaywarrick/Documents/MMB/Projects/Dominique/Dom1to3_r1234')
dir.create('G:/Data Tables/Dom1to3_r1234')
for(feature in rfImp$name[1:100])
{
temp <- gsub(".","_",feature, fixed=T)
#filePath <- file.path('/Users/jaywarrick/Documents/MMB/Projects/Dominique/Dom2',paste0(i, "_", temp,'.pdf'))
filePath <- file.path('G:/Data Tables/Dom1to3_r1234',paste0(i, "_", temp,'.pdf'))
pdf(file=filePath, width=6, height=5)
plotHist(shinyData,feature)
dev.off()
i <- i + 1
}
plotHist(x3,'ZernikeDot40_THISwSEC_WholeCell_390X440_dot_648X684')
plotHist(x3,'ZernikeDot20_THISwSEC_WholeCell_390X440_dot_648X684')
plotHist(x3,'Geometric.SizeIterable_WholeCell_None')
plotHist(x3,'Geometric.SizeIterable_Nuc_None')
plotHist(x3,'Stats.Sum_Nuc_390X440')
plotHist(x3,'Stats.Sum_WholeCell_390X440')
plotHist(x3,'NCR_Sum')
plotHist(x3,'NCR_Mean')
# Plot N/C
x1b[MaskChannel=='Nuc' & ImageChannel=='648 X 684' & Measurement=='Stats.Sum']/x1b[MaskChannel=='Cyt' & ImageChannel=='648 X 684' & Measurement=='Stats.Sum']
x3[,NCR_Sum:=Stats.Sum_Nuc_648X684/Stats.Sum_Cyt_648X684,by='Class']
x3[,NCR_Mean:=Stats.Mean_Nuc_648X684/Stats.Mean_Cyt_648X684,by='Class']
x2[MaskChannel=='Nuc',mean(Geometric.SizeIterable, na.rm=T),by='Class']
x2[MaskChannel=='WholeCell',mean(Geometric.SizeIterable, na.rm=T),by='Class']
x3[,mean(Stats.Sum_Nuc_390X440, na.rm=T),by='Class']
##### DEBUGGING & TESTS #####
names(dataToTest)[as.logical(as.vector(dataToTest[,lapply(.SD, function(x){length(which(is.na(x)))>1})]))]
names(dataToTest)[as.logical(as.vector(dataToTest[,lapply(.SD, function(x){length(which(is.null(x)))>1})]))]
#### Test for index issue #####
order1 <- data.table(ImRow=0:5, ImCol=rep(0:5, each=6), index1=0:35)
order2 <- data.table(ImRow=rep(0:5,each=6), ImCol=0:5, index2=0:35)
order3 <- data.table(ImRow=0:5, ImCol=rep(0:5, each=6), index3=c(0:5,rev(6:11),12:17,rev(18:23),24:29,rev(30:35)))
duh <- data.table(ImRow=c(1,3,5,0),ImCol=c(0,5,4,1), other=c('a','b','c','d'))
shinyData <- merge(order1, shinyData, by=c('ImRow','ImCol'), all=FALSE)
shinyData <- merge(order2, shinyData, by=c('ImRow','ImCol'), all=FALSE)
shinyData <- merge(order3, shinyData, by=c('ImRow','ImCol'), all=FALSE)
library(Hmisc)
mt <- shinyData[Class=='MT']
wt <- shinyData[Class=='WT']
X <- mt[,c('index1','index2','index3'),with=F]
yNames <- getNumericCols(mt)[!(getNumericCols(mt) %in% c('index1','index2','index3'))]
Y <- mt[,yNames, with=F]
temp <- as.data.frame(cor(data.frame(X),data.frame(Y)))
cors <- data.frame(var=names(temp), cor1=as.numeric(as.vector(temp[1,])), cor2=as.numeric(as.vector(temp[2,])), cor3=as.numeric(as.vector(temp[3,])))
cors1 <- cors[order(abs(cors$cor1), decreasing=TRUE),][,c('var','cor1')]
cors2 <- cors[order(abs(cors$cor2), decreasing=TRUE),][,c('var','cor2')]
cors3 <- cors[order(abs(cors$cor3), decreasing=TRUE),][,c('var','cor3')]
nameToGet1 <- as.character(cors1$var[2])
nameToGet2 <- as.character(cors2$var[2])
nameToGet3 <- as.character(cors3$var[2])
plot(mt[,get('index1')], mt[,get(nameToGet1)])
plot(mt[,get('index2')], mt[,get(nameToGet2)])
plot(mt[,get('index3')], mt[,get(nameToGet3)])
# Yay, no trend!
##### Debug stuff #####
x2[Measurement=='HistogramBin_13' & MaskChannel=='WholeCell']
MADs <- shinyData[,lapply(.SD, mad, na.rm=TRUE), .SDcols=getNumericCols(shinyData)]
SDs <- shinyData[,lapply(.SD, sd, na.rm=TRUE), .SDcols=getNumericCols(shinyData)]
SDs$net.imagej.ops.Ops.Stats.StdDev_WholeCell_650X705M
# Probability of flipping 57 heads out of 100 flips given a fair (0.5 fraction) coin is
1-pbinom(0.51*15000,15000,0.50)
qbinom(c(0.025,0.975),100,0.5)
duh <- copy(x1a)
fixColNames(duh)
fixNames(duh, col=c('MaskChannel','ImageChannel','Measurement'))
replaceStringInAllRowsOfCol(duh,'Measurement',"net.imagej.ops.Ops.","")
duh$Measurement <- paste(duh$Measurement, duh$MaskChannel, duh$ImageChannel, sep='_')
duh[,c('MaskChannel','ImageChannel'):=NULL]
duh <- getWideTable(duh)
hist(duh$Stats.Mean_WholeCell_485X525M, breaks=100)
plot_ly(mode='markers',x=duh$Stats.Mean_WholeCell_485X525M-500, y=duh$Stats.Mean_WholeCell_395X455M-500)
layout(p, xaxis = list(type = "log"),
yaxis = list(type = "log"), hovermode="closest")
shinyData <- duh
shinyData$Class <- 'MT'
shinyData[800:1600]$Class <- 'WT'