Using RTCGA.PANCAN12 package to compare time to death for selected tumor types

Przemyslaw Biecek

2017-04-26

RTCGA.PANCAN12 package

You need RTCGA.PANCAN12 package to use PANCAN12 data from Cancer Genome Browser.

## try http:// if https:// URLs are not supported
source("https://bioconductor.org/biocLite.R")
biocLite("RTCGA.PANCAN12")
# or try devel version
require(devtools)
if (!require(RTCGA.PANCAN12)) {
    install_github("RTCGA/RTCGA.PANCAN12")
    require(RTCGA.PANCAN12)
}
# or if you have RTCGA package then simpler code is
RTCGA::installTCGA('RTCGA.PANCAN12')

Expression

Expression data is divided and stored in two datasets, because it was too big to fit GitHub limits. So first you need to do is to bind both data sets into one.

expression.cb <- rbind(expression.cb1, expression.cb2)

Now, let’s see where is expression for MDM2 and let’s create a dataset with single column single - MDM2 expression.

grep(expression.cb[,1], pattern="MDM2")

MDM2 <- expression.cb[8467,-1]
MDM2v <- t(MDM2)

Mutations

Mutations are stored in mutation.cb and they are coded 0/1. Let’s gather all mutations for TP53 and then let’s create a data frame with single column - only for TP53?

grep(mutation.cb[,1], pattern="TP53$", value = FALSE)

TP53 <- mutation.cb[18475,-1]
TP53v <- t(TP53)

Merging

From clinical data we are going to extract following variables: - X_cohort (cancer type) - X_EVENT (0/1) - X_TIME_TO_EVENT (in days) - X_PANCAN_UNC_RNAseq_PANCAN_K16 (cancer’s subtypes)

And then we merge clinical, expression and mutation datasets. Not that it requires some data cleaning (in clinical data _ is used as a separator while in expression and mutation it’s .).

dfC <- data.frame(names=gsub(clinical.cb[,1], pattern="-", replacement="."), clinical.cb[,c("X_cohort","X_TIME_TO_EVENT","X_EVENT","X_PANCAN_UNC_RNAseq_PANCAN_K16")])
dfT <- data.frame(names=rownames(TP53v), vT = TP53v)
dfM <- data.frame(names=rownames(MDM2v), vM = MDM2v)
dfTMC <- merge(merge(dfT, dfM), dfC)
colnames(dfTMC) = c("names", "TP53", "MDM2", "cohort","TIME_TO_EVENT","EVENT","PANCAN_UNC_RNAseq_PANCAN_K16")
dfTMC$TP53 <- factor(dfTMC$TP53)

# only primary tumor
# (removed because of Leukemia)
# dfTMC <- dfTMC[grep(dfTMC$names, pattern="01$"),]

MDM2 and TP53

First let’s see the MDM2 expression along cancers.

library(ggplot2)
quantile <- stats::quantile
ggplot(dfTMC, aes(x=cohort, y=MDM2)) + geom_boxplot() + theme_bw() + coord_flip() + ylab("")

And let’s see the fraction of TP53 in cancers

ggplot(dfTMC, aes(x=cohort, fill=TP53)) + geom_bar() + theme_bw() + coord_flip() + ylab("")

And how many cases for particular cancer types?

sort(table(dfTMC$cohort))

Survival in different cancer types given MDM2 and TP53

Let’s dichotomize MDM2 into two groups with the cutoff =0 (almost median).

dfTMC$MDM2b <- cut(dfTMC$MDM2, c(-100,0,100), labels=c("low", "high"))

Number of cases

library(dplyr)
library(tidyr)
dfTMC %>% 
  group_by(MDM2b, TP53, cohort) %>%
  summarize(count=n()) %>%
  unite(TP53_MDM2, TP53, MDM2b) %>%
  spread(TP53_MDM2, count, fill = 0)

For four cancers with largest number of cases let’s see Kaplan Meier curves divided into MDM2/TP53 groups.

Comments:

For Breast there is a very nice relation between high MDM2/mutated TP53 and survival. Outcomes for first 5 years are much worse.

For Kidney mutations in TP53 are uncommon, but both groups have bad prognosis.

For Head and Neck mutations in TP53 have bad prognosis and id MDM2 is high the prognosis is even worse.

For Endomeroid the interesting group is very small, for Lung there is no clear pattern. Remaining cancers are very small.

library(survey)
library(scales)
library(survMisc)

# cancer = "TCGA Breast Cancer"
cancers <- names(sort(-table(dfTMC$cohort)))

for (cancer in cancers[1:11]) {
  survp <- survfit(Surv(TIME_TO_EVENT/356,EVENT)~TP53+MDM2b, data=dfTMC, subset=cohort == cancer)
  pl <- autoplot(survp, title = "")$plot + theme_bw() + scale_x_continuous(limits=c(0,10), breaks=0:10) + ggtitle(cancer) + scale_y_continuous(labels = percent, limits=c(0,1))
  cat(cancer,"\n")
  plot(pl)
}