ziesel.hierarchy.Rmd

---
title: "Jamboree 2: SpaceTx task: Ziesel et al. hierarchy"
output: html_notebook
---

Load Ziesel et al. mRNA data matrix
```{r}
require(data.table)
scrna.dat <- as.data.frame(fread("expression_mRNA_17-Aug-2014.txt",skip=11,header=F))
# extract gene names
rownames(scrna.dat) <- scrna.dat[,1];
# remove first two (non-count) columns
scrna.dat <- scrna.dat[,-c(1,2)]
# dimensions of the final matrix
dim(scrna.dat)
```

Read in cell metadata
```{r}
scrna.mdat <- t(read.table("expression_mRNA_17-Aug-2014.txt",sep='\t',nrows=3,skip=7))
colnames(scrna.mdat) <- scrna.mdat[2,]; 
scrna.mdat <- data.frame(scrna.mdat[-c(1:2),],stringsAsFactors=F)
head(scrna.mdat)
```

```{r}
# assign cell names to the data matrix
colnames(scrna.dat) <- scrna.mdat[,'cell_id']
# level 2 class counts
table(scrna.mdat$level2class)
```


We'll build hierarchy of clusters (suing 'complete' linkage) by pooling count data for each level 2 cluster, and calculating similarity between clusters using 1-r (Pearson correlation):

```{r}
# pool read counts across clusters
scrna.pooled <- do.call(cbind,tapply(1:ncol(scrna.dat), as.factor(scrna.mdat$level2class), function(ii) {
  rowSums(scrna.dat[,ii])
}))

# normalize and log transform
lib.size.scale <- 1e4;
scrna.pooled.norm <- log( t(t(scrna.pooled)/colSums(scrna.pooled)) * lib.size.scale +1);

# calculate correlation distance
scrna.pooled.norm.dist <- 1-cor(scrna.pooled.norm)

# cluster
scrna.pooled.hc <- hclust(as.dist(scrna.pooled.norm.dist), method='complete')

# show clustering
plot(scrna.pooled.hc)
```