Normalizing mutation counts by Gene size

[Kotoro]

What would the recommended standard be for doing a normalization to eliminate or reduce the impact of gene size, and where should I go to obtain the data to do this?

I am working with TCGA MAF files so I have entrez IDs and Hugo names for the genes. I would like to normalize the counts according to the size of the gene.

I THINK tcga only includes mutations that were within the expressed sequences, but I would have to double check to be sure. (if any of you know the answer for certain that would be appreciated.)

Would the answer make a difference as to whether I should normalize only according to total number of nucleotides in exon sequences (and leave out the intron lengths?), or should I go for total gene start-stop length anyway?

The data is Human data. To be clear, by counts I mean # of mutations per gene, per sample or grouping of samples.

[Kotoro]

I'm surprised. I figured this one would be easy.

[dpryan]

I guess no one felt like answering. Normalize by total exonic length, thereby excluding introns. In theory, one could add a few bases per intron to account for finding splice-site mutations, but the difference due to that will be miniscule. You can get that from Ensembl/gencode/UCSC. Just parse the GTF file (it might be convenient to do this in R. I wouldn't be surprised if the UCSC table browser already has this info in some random table as well.

[Kotoro]

Thank you for your response. I figured that exon-length sums was the best way to go for this, but wasn't sure. Thanks. I've pulled down some stuff from UCSC and am starting to look at my options in terms of the table browser for gene lists.

I don't really KNOW R, though I could easily find a library in either perl or python to use to compute the numbers relatively quickly. (something this simple doesn't warrant a C program I would think)

Any idea which of the known human genes tracks in the genome browser would be best to pair up with the TCGA data? (their MAF files use Hugo gene symbol and Entrez Gene ID identifiers).

[dpryan]

Yup, a little python or perl script would make more sense than something in C. I expect that the refseq track matches the best, since that would likely use Entrez IDs already.