# How to read and interpret a gene expression quantification file?

I have a gene expression quantification file from TCGA that contains the following lines:

ENSG00000242268.2   591.041000514
ENSG00000270112.3   0.0
ENSG00000167578.15  62780.6543066
ENSG00000273842.1   0.0
ENSG00000078237.5   36230.832883
ENSG00000146083.10  189653.152706
ENSG00000225275.4   0.0
ENSG00000158486.12  420.761140072
ENSG00000198242.12  2914738.3675
ENSG00000259883.1   1632.83700531
ENSG00000231981.3   0.0
ENSG00000269475.2   0.0
ENSG00000201788.1   0.0
ENSG00000134108.11  925529.547944
ENSG00000263089.1   2646.63769677
ENSG00000172137.17  23162.6989867
ENSG00000167700.7   291192.25157

1. What is the .number that are added to the Gene e.g. the .2 in ENSG00000242268.2

2. Why isn't the quantified value an integer; what does 591.041000514 mean?

1. The first column contains Ensembl gene identifiers, and the suffix is a version number that can be used to track changes to the gene annotations over time. From the Ensembl Stable IDs documentation:

Ensembl annotation uses a system of stable IDs that have prefixes based on the species scientific name plus the feature type, followed by a series of digits and a version e.g. ENSG00000139618.1. The version may be omitted.

2. Following the first link you provided leads to a page with details for the file 2edcaaa7-63b4-40b4-abbe-5d7a84012e60.FPKM-UQ.txt.gz. The first thing that caught my eye about this filename was FPKM, or "fragments per kilobase of exon per million reads", which is a commonly used unit of RNA expression. Since these are not raw read counts, there's no expectation that these values should be integers.

The best explanation I've seen of FPKM comes from a blog post written by Harold Pimentel of kallisto and sleuth fame. From the blog post:

The interpretation of FPKM is as follows: if you were to sequence this pool of RNA again, you expect to see FPKM_i fragments for each thousand bases in the feature for every N/10^6 fragments you’ve sequenced. It’s basically just the rate of fragments per base multiplied by a big number (proportional to the number of fragments you sequenced) to make it more convenient.

More generally though, even when FPKM is not the unit of expression abundance used, most quantification methods and associated units of expression will not produce integer estimates.

• Tools such as DESeq2 are able to generate other expression statistics that are similar to, but not the same as, FPKM. I'm not sure if it makes sense to answer this for the specific question (i.e. because of FPKM normalisation), or in the more general case, where I would say that the values are floating point due to various different normalisation procedures (e.g. correction for transcript length, sequencing depth, shot noise, confounding covariates).
– gringer
Jun 8 '17 at 5:46