When working with Bioconductor and the curatedOvarianData dataset, what are the unit values of the expression data?

Question

I am wondering how to interpret values inside the gene expression levels contained within Bioconductor. For example, if we have the following commands via R, we get:

source("https://bioconductor.org/biocLite.R") 
biocLite("curatedOvarianData")
library(curatedOvarianData)
data(TCGA_eset)
head(exprs(TCGA_eset)["CXCL12",])

TCGA.20.0987 TCGA.23.1031 TCGA.24.0979 TCGA.23.1117 TCGA.23.1021 TCGA.04.1337 
    5.437923     6.307250     4.193748     9.071293     5.101532     6.697770 

The above calls out gene expression levels for the gene CXCL12. However, I am wondering how to interpret the values 5.437923, 6.307250, etc.

I know they are expression levels, but do they come with units? Do larger values mean greater cancer possibility?

Answer 1

These values represent the expression of a particular gene in each sample. While expression of an immune system gene like CXLC12 might be associated with the aggressiveness of a cancer, that's not going to be the case for the majority of genes.

Looking at the numbers, these look like they are probably log$_2$-transformed values. In order to know for sure, plotting a density distribution of the dataset is a good idea:

plot(density(exprs(TCGA_eset)));

If the values are in the range of around 0..15, then it'll be log$_2$-transformed values, whereas if they're in the range of 0..100000, it'll be "raw" expression values. The log-transformed values are preferred for data analysis because gene expression tends to be close to a normal distribution in log-space.

These numbers are best treated as relative counts (i.e. the most appropriate unit is "expression"), as variation in the preparation of the sample (or the sequencer, or the microarray) can influence the results that come out the other end. Converting them to something more meaningful (e.g. "transcripts per cell") requires normalising against a known reference sample.

Answer 2

The values are rma normalized expression values from affy microarrays. They are indeed log2 transformed (as @gringer already expected). You can find more info about the data here, under heading TCGA_eset.

To get more info about rma, check this out.

Larger values are not necessarily greater cancer possibility, I wish biology was that simple.