Simon DESeq.Rmd

DE via Pseudobulk
=================

Simon Anders, 2019-07-30

Using the data from the ASD paper, this short notebook shows how a DE analysis can be done using pseudobulks: For a given cell sub-population 
(here used as example: the L2/3 neurons), add up the counts over all cells in each sample, then perform a normal DESeq2 analysis on that.

```{r}
suppressPackageStartupMessages({
library( tidyverse )
library( Matrix )
library( DESeq2 )
})
```

Load ASD data:
```{r}
path <- "~/Desktop/ASD/rawMatrix/"
cellinfo <- read.delim( file.path( path, "meta.txt" ), stringsAsFactors=FALSE )
counts <- readMM( file.path( path, "matrix.mtx" ) )
rownames(counts) <- read.delim( file.path( path, "genes.tsv" ), header=FALSE )$V2
colnames(counts) <- readLines( file.path( path, "barcodes.tsv" ) )
head(cellinfo)
```

Convert counts from a coordinate-sparse to a column-sparse matrix. This speeds up the calculation of the pseudobulks (but does require lots of memory here).

```{r}
counts <- as( counts, "dgCMatrix" )
counts[ 1:5, 1:5 ]
```

Extract sample information from 'meta' table:
```{r}
sampleTable <-
  cellinfo %>% select( sample : RNA.Integrity.Number ) %>% unique
sampleTable
```

Make pseudobulk for L2/3 cells by simply summing up L2/3 cells for each sample
```{r}
pseudobulk_L23 <-
  sapply( sampleTable$sample, function(s)
    rowSums( counts[ , cellinfo$sample == s & cellinfo$cluster=="L2/3", drop=FALSE ] ) )
pseudobulk_L23[ 1:5, 1:5 ]
```

Run DESeq on the pseudobulk. 

```{r}
dds <- DESeqDataSetFromMatrix( pseudobulk_L23, sampleTable, ~ age + sex + region + diagnosis )
dds <- DESeq( dds )
res <- results( dds )
```

Show top hits
```{r}
res[ order(res$padj), ]

```

How many hits at 10% FDR?

```{r}
table( res$padj < .1 )
```

Session info

```{r}
sessionInfo()
```