Using ASCAT for multi-sample analyses
We start by loading the ASCAT package.
Next we load the data.
ascat.bcMulti <- ascat.loadData(
Tumor_LogR_file = system.file("extdata", "tumour.logR.txt", package="ASCAT"),
Tumor_BAF_file = system.file("extdata", "tumour.BAF.txt", package="ASCAT"),
Germline_LogR_file = system.file("extdata", "singlenormal.logR.txt", package="ASCAT"),
Germline_BAF_file = system.file("extdata", "singlenormal.BAF.txt", package="ASCAT"))
## [1] Reading Tumor LogR data...
## [1] Reading Tumor BAF data...
## [1] Reading Germline LogR data...
## [1] Reading Germline BAF data...
## [1] Registering SNP locations...
## [1] Splitting genome in distinct chunks...
Both Tumor_LogR_file and Tumor_BAF_file are
expected to contain a column for each of the samples to analyse.
head(ascat.bcMulti$Tumor_LogR)
## S1 S2
## SNP1 0.03615 -1.03950
## SNP2 0.14998 -0.79433
## SNP3 -0.00891 -0.76137
## SNP4 0.40188 -0.67521
## SNP5 0.14902 -0.72980
## SNP6 0.24118 -1.11302
head(ascat.bcMulti$Tumor_BAF)
## S1 S2
## SNP1 0.51596 0.99262
## SNP2 0.67903 0.00255
## SNP3 1.00000 1.00000
## SNP4 0.00000 0.00000
## SNP5 1.00000 1.00000
## SNP6 0.45572 0.04925
The next step is to run the segmentation. When analysing
phylogenetically related samples, it is expected that some of the copy
number segment boundaries are shared between samples. In this case a
joint segmentation of all samples is recommended. The synthetic data set
used in this example was also simulated with partly shared segment
boundaries. The ground truth copy number plots of the two samples we are
going to analyse are shown in the following plots.
The multi-sample segmentation algorithm can be run using the function
ascat.asmultipcf.
ascat.bcMulti <- ascat.asmultipcf(ascat.bcMulti,penalty = 5,out.dir=NA)
## [1] "Segmentlength 5"
Finally ASCAT can be run on the segmented data set.
ascat.outputMulti = ascat.runAscat(ascat.bcMulti)
Comparison with single sample segmentation
Finally, we compare our result to that of standard single sample
segmentation using ascat.aspcf.
ascat.bc = ascat.loadData(system.file("extdata", "tumour.logR.txt", package="ASCAT"),
system.file("extdata", "tumour.BAF.txt", package="ASCAT"),
system.file("extdata", "normal.logR.txt", package="ASCAT"),
system.file("extdata", "normal.BAF.txt", package="ASCAT"))
## [1] Reading Tumor LogR data...
## [1] Reading Tumor BAF data...
## [1] Reading Germline LogR data...
## [1] Reading Germline BAF data...
## [1] Registering SNP locations...
## [1] Splitting genome in distinct chunks...
ascat.bc = ascat.aspcf(ascat.bc,penalty = 25,out.dir=NA)
## [1] Sample S1 (1/2)
## [1] Sample S2 (2/2)
Note that in the single-sample case the same segmentation sensitivity
is achieved with a higher penalty parameter compared to the multi-sample
case. This means, when switching from single- to multi-sample
segmentation, the penalty parameter needs to be lowered to maintain a
similar sensitivity.
We plot the segment boundaries inferred for each of the two samples
by multi- and single-sample segmentation.
plot.segments(v1=cumsum(rle(ascat.bc$Tumor_LogR_segmented[,1])$lengths),
v2=cumsum(rle(ascat.bc$Tumor_LogR_segmented[,2])$lengths),
main="Single-sample segmentation")
plot.segments(v1=cumsum(rle(ascat.bcMulti$Tumor_LogR_segmented[,1])$lengths),
v2=cumsum(rle(ascat.bcMulti$Tumor_LogR_segmented[,2])$lengths),
main="Multi-sample segmentation")
In case of single-sample segmentation the inferred positions of most of
the shared segment boundaries vary slightly between the two samples,
whereas the multi-sample segmentation infers a common breakpoint when
there is no significant evidence that the boundaries differ between
samples.
##Comparison with another multi-sample copy number method
In order to compare asmultipcf segmentation to other copy number
inference methods, we ran ASCAT on two samples from two patients from a
metastatic prostate cancer study (Gundem et al).
For the same study, copy number profiles are available from HATCHet (github repository;
publication).
Loading all neccessary packages.
library(ggplot2)
library(plyr)
Because HATCHet can model a mixture of copy number states but ASCAT
only detects the major clone, we need to define the major clone in the
HATCHet data.
getMajorClone<-function(row){
maj<-names(which.max(row[c(8,10,12)]))
maj.cn<-row[(match(maj,names(row))-1)]
return(maj.cn)
}
Next we define a function to load HATCHet data, define the major
clone for each segment, i.e. row, and split copy number annotation from
“a|b” format into two columns.
readHatchetFile<-function(path){
hatchet.raw<-read.table(path, header=F, sep="\t", stringsAsFactors = F)
##find major clone
major.cn<-apply(hatchet.raw, 1, function(z) getMajorClone(z))
hatchet.raw$major.cn<-major.cn
hatchet.short<-hatchet.raw[c(1:4,13)]
names(hatchet.short)<-c("chr","start","end","sample","hatchet.cn")
major.cn<-unlist(lapply(strsplit(hatchet.short$hatchet.cn, "\\|"), function(z) z[1]))
minor.cn<-unlist(lapply(strsplit(hatchet.short$hatchet.cn, "\\|"), function(z) z[2]))
hatchet.short$major.cn<-major.cn
hatchet.short$minor.cn<-minor.cn
return(hatchet.short[-5])
}
In order to provide a quantitative measure of how much ASCAT using
asmultipcf and HATCHet agree with regard to the sample segmentation, we
analysed what fraction of asmultipcf breakpoints has a corresponding
HATCHet breakpoint. First, we centred the ASCAT breakpoints in the
middle of two segments.
returnBreakpoints<-function(df.chr){
bp.df<-NULL
df.sorted<-df.chr[order(df.chr$start),]
for(i in 1:(nrow(df.sorted)-1)){
first<-df.sorted[i,]
second<-df.sorted[(i+1),]
##verify real cn bp
if(first$major.cn==second$major.cn & first$minor.cn==second$minor.cn){
print("Consecutive rows have identical CN!")
}
else{
bp.df<-rbind(bp.df, data.frame(bp.location=(first$end+((second$start-first$end)/2))))
}
}
return(bp.df)
}
HATCHet output is provided for 50kb bins and not specifically
segmented. Hence, in order to compare consecutive segments with the same
major clonal copy number, we are merging neighbouring segments with
matching allele specific copy number values. Then we calculate the
distance of the closest HATCHet breakpoint for every asmultipcf
breakpoint.
findMatchingBPs<-function(ascat.bps, hatchet.segs){
##limit hatchet df to true CN segs
hatchet.true<-NULL
hatchet.sorted<-hatchet.segs[order(hatchet.segs$start),]
rles<-rle(paste0(hatchet.sorted$major.cn,"_",hatchet.sorted$minor.cn))$lengths
idx<-1
for(k in rles){
hatchet.true<-rbind(hatchet.true, data.frame(start=hatchet.sorted[idx, "start"], end=hatchet.sorted[(idx+k-1), "end"]))
idx<-(idx+k)
}
ascat.flag<-NULL
for(i in 1:nrow(ascat.bps)){
ascat.bp<-ascat.bps[i,"bp.location"]
min.dist<-min(c(abs(ascat.bp-hatchet.true$end),abs(ascat.bp-hatchet.true$start)))
ascat.flag<-rbind(ascat.flag, data.frame(bkp=ascat.bp, min.dist=min.dist))
}
return(ascat.flag)
}
This is the main function visualising the copy number of asmultipcf
and HATCHet for all four analysed samples and calculates the fraction of
ASCAT breakpoints with a HATCHet breakpoint fewer than 50kb bases away
(size of HATCHet bins).
compareCN<-function(cn.ascat, cn.hatchet){
##split into samples
shared.samples<-intersect(cn.ascat$sample, cn.hatchet$sample)
for(s in shared.samples){
ascat.sub<-cn.ascat[cn.ascat$sample==s,]
hatchet.sub<-cn.hatchet[cn.hatchet$sample==s,]
##move values slightly for plotting
ascat.plot<- ascat.sub
hatchet.plot<- hatchet.sub
ascat.plot$major.cn<-(as.numeric(ascat.plot$major.cn)+0.1)
hatchet.plot$major.cn<-(as.numeric(hatchet.plot$major.cn)-0.1)
##visualise
joint.df<-rbind(ascat.plot,hatchet.plot)
joint.df$chr<-factor(joint.df$chr, levels=c(1:22,"X","Y"))
p1<-ggplot(joint.df, aes(x=start, y=(as.numeric(major.cn)+as.numeric(minor.cn)), xend=end, yend=(as.numeric(major.cn)+as.numeric(minor.cn)), col=as.factor(method)))+
geom_segment()+
theme(axis.text.x=element_text(angle=45,hjust=1), panel.background = element_blank(), panel.grid.major = element_line(color="grey80"))+
ggtitle(paste0("Sample ",s))+
scale_color_manual(values=c("cyan4","coral3"), name="")+
scale_x_continuous(name="")+
scale_y_continuous(name="Copy number", limits=(c(0,15)))+
facet_wrap(.~chr, scales = "free_x", nrow = 3)
print(p1)
##give quality in measure: what fraction of ASCAT bp has hatchet bp in < threshold vincinity?
bps<-ddply(ascat.sub, .(chr), function(z) returnBreakpoints(z))
bp.list<-split(bps, bps$chr)
bkp.list<-NULL
for(c in names(bp.list)){
if(c!="X"){
bkp.list<-rbind(bkp.list,findMatchingBPs(bp.list[[c]], hatchet.sub[hatchet.sub$chr==c,]))
}
}
##use bin size of HATCHet as threshold
threshold=50000
bkp.list$match<-ifelse(bkp.list$min.dist<threshold,T,F)
print(s)
print(prop.table(table(bkp.list$match)))
}
}
Now we can carry out the comparison. Load the HATCHet segmentation
for patient A32.
path<-"https://raw.githubusercontent.com/raphael-group/hatchet-paper/master/cancer/prostate/A32/best.seg.ucn"
hatchet.seg<-readHatchetFile(path)
hatchet.seg$method<-"HATCHet"
hatchet.seg$sample<-gsub("-","",hatchet.seg$sample)
Now we can load the asmultipcf results for patient A32.
ascat.seg<-read.table(system.file("extdata", "A32.fast.segments.txt", package="ASCAT"), header=T, sep="\t", stringsAsFactors = F)
ascat.seg$tmp<-unlist(lapply(strsplit(ascat.seg$sample, "-"), function(z) z[1]))
ascat.seg<-ascat.seg[-1]
names(ascat.seg)<-c("chr","start","end","major.cn","minor.cn","sample")
ascat.seg$method<-"ASCAT"
Finally, we can create the visual and segmentation comparison between
the copy number of asmultipcf and HATCHet.
compareCN(ascat.seg, hatchet.seg)
## [1] "A32C"
##
## FALSE TRUE
## 0.3591331 0.6408669
## [1] "A32A"
##
## FALSE TRUE
## 0.4606061 0.5393939
So for both samples of patient A32, between 54% and 64% of asmultipcf
breakpoints have a corresponding HATCHet breakpoint.
We can now repeat the same analysis for patient A17.
path<-"https://raw.githubusercontent.com/raphael-group/hatchet-paper/master/cancer/prostate/A17/best.seg.ucn"
hatchet.seg<-readHatchetFile(path)
hatchet.seg$method<-"HATCHet"
ascat.seg<-read.table(system.file("extdata", "A17.fast.segments.txt", package="ASCAT"), header=T, sep="\t", stringsAsFactors = F)
ascat.seg$tmp<-unlist(lapply(strsplit(ascat.seg$sample, "-"), function(z) z[1]))
ascat.seg<-ascat.seg[-1]
names(ascat.seg)<-c("chr","start","end","major.cn","minor.cn","sample")
ascat.seg$method<-"ASCAT"
compareCN(ascat.seg, hatchet.seg)
## [1] "A17F"
##
## FALSE TRUE
## 0.2039801 0.7960199
## [1] "A17A"
##
## FALSE TRUE
## 0.1955017 0.8044983
For both samples of patient A17, around 80% of asmultipcf breakpoints
have a corresponding HATCHet breakpoint.