purity.Rmd

---
title: "tumor purity estimation"
author: "Martin Sill"
date: "3/16/2018"
output: html_document
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
```

##### TCGA data
ABSOLUTE purity values for the TCGA data analysed in 
"Molecular Profiling Reveals Biologically Discrete Subsets and Pathways of Progression in Diffuse Glioma" (Ceccarelli et al., 2016)
are available at <http://www.cell.com/cms/attachment/2045372863/2056783242/mmc2.xlsx>.
The normalized methylation data is available at <https://portal.gdc.cancer.gov/>


```{r cars}
rm(list=ls())
library(randomForest)

load("tcga_ceccarelli_2016.RData")
ls()
betas[1:5,1:5]
df$Case[1:5]
df$ABSOLUTE.purity[1:5]

# feature selection  
rf <- randomForest(df$ABSOLUTE.purity,x=t(betas),ntree = 100)
oo <- order(rf$importance,decreasing = T)
# train final rf model
rf <- randomForest(df$ABSOLUTE.purity,x=t(betas[oo[1:1000],]),ntree = 500)  

rf
```

```{r pressure, echo=FALSE}
# plot out of bag error
plot(rf$y,y=rf$predicted,pch=16,xlab="Absolute purity",ylab="RF predicted (oob)",ylim=c(0,1),xlim=c(0,1))
abline(0,1,col="red")
legend("topleft",legend=paste0("MSE=",round(rf$mse[500],5)))
```