data(analyzed)
context('re-use.groups with FACS')
onwhat='Expression'
clusterby='MDS'
mds.type='PCA'
move.neg <- TRUE
plot.neg <- TRUE
beanplots = TRUE
plotsvg = 0
zscoredVioplot = 1
cmethod='ward.D'
LLEK='2'
ctype= 'hierarchical clust'
groups.n <- 5
data@outpath = tempdir()
data2 <- analyse.data (
data,
groups.n=4,
onwhat='Expression',
clusterby='MDS',
mds.type='PCA',
cmethod='ward.D',
LLEK='2',
ctype= 'hierarchical clust',
zscoredVioplot = zscoredVioplot,
move.neg = move.neg,
plot.neg=plot.neg,
beanplots=beanplots,
useGrouping='ArrayID',
plotsvg = 0
)
expect_that( colnames(data2@samples) , equals(colnames(data@samples)) )
data2 <- group_1D (data, 'CD41.Alexa.Fluor.488.A', c(0.5, 1, 1.5, 2, 2.5, 3 ) )
expect_that(
colnames(data2@samples) ,
equals(c(colnames(data@samples) , 'CD41.Alexa.Fluor.488.A 1D Group' ) )
)
## check that the values are also OK
expect_that ( all( data2@facs[ which(data2@samples[,'CD41.Alexa.Fluor.488.A 1D Group'] == 1), 'CD41.Alexa.Fluor.488.A' ] <= 0.5), is_true() )
expect_that ( all( data2@facs[ which(data2@samples[,'CD41.Alexa.Fluor.488.A 1D Group'] == 2), 'CD41.Alexa.Fluor.488.A' ] <= 1), is_true() )
expect_that ( all( data2@facs[ which(data2@samples[,'CD41.Alexa.Fluor.488.A 1D Group'] == 3), 'CD41.Alexa.Fluor.488.A' ] <= 1.5), is_true() )
expect_that ( all( data2@facs[ which(data2@samples[,'CD41.Alexa.Fluor.488.A 1D Group'] == 4), 'CD41.Alexa.Fluor.488.A' ] <= 2), is_true() )
expect_that ( all( data2@facs[ which(data2@samples[,'CD41.Alexa.Fluor.488.A 1D Group'] == 5), 'CD41.Alexa.Fluor.488.A' ] <= 2.5), is_true() )
expect_that ( all( data2@facs[ which(data2@samples[,'CD41.Alexa.Fluor.488.A 1D Group'] == 6), 'CD41.Alexa.Fluor.488.A' ] <= 3), is_true() )
expect_that ( all( data2@facs[ which(data2@samples[,'CD41.Alexa.Fluor.488.A 1D Group'] == 7), 'CD41.Alexa.Fluor.488.A' ] >= 3), is_true() )
data2 <- group_1D (data2, 'Actb', c( 26,30,32))
expect_that ( all( data2@data[ which(data2@samples[,'Actb 1D Group'] == 1), 'Actb' ] <= 26), is_true() )
expect_that ( all( data2@data[ which(data2@samples[,'Actb 1D Group'] == 2), 'Actb' ] <= 30), is_true() )
expect_that ( all( data2@data[ which(data2@samples[,'Actb 1D Group'] == 3), 'Actb' ] <= 32), is_true() )
expect_that ( all( data2@data[ which(data2@samples[,'Actb 1D Group'] == 4), 'Actb' ] >= 32), is_true() )
gr_old <- split( data2@samples[,1], data2@samples[,4])
l <- list()
expect_that(names(gr_old), equals( paste(1:7)) )
for ( i in c( '2','5','4','1','3','6','7') ) { l[[length(l)+1]] = gr_old[[i]]}
names(l) <- 1:7
data2@samples$reorderedGroup <- data2@samples[,4]
data2 <- regroup(data2, group2sample=l, name = 'reorderedGroup' )
t <- table (data2@samples[,c(4,5)])
expect_that(t[c(2,7+5,7*2+4,7*3+1,7*4+3,7*5+6,7*6+7)],equals( c(12,62,161,9,33,4,1)) )
data2 <- group_on_strings( data2, strings=c( ' P1', ' P2', ' P3' ) )
t <- table (data2@samples[,c(2,6)])
expect_that(t[c(1,2,3,4,8,12)],equals( c(5,5,3,89,89,91)) )
colColors = list(
ArrayID=rainbow(max(data2@samples$ArrayID)),
'auto_clusters.1' = rainbow(max(data2@samples$'auto_clusters.1')),
'CD41.Alexa.Fluor.488.A 1D Group' = gray.colors( max(data2@samples$'CD41.Alexa.Fluor.488.A 1D Group'))
)
complexHeatmap( data2, ofile= 'notUsed', colGroups= c('ArrayID', 'auto_clusters.1', 'CD41.Alexa.Fluor.488.A 1D Group' ),colColors= colColors)
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.