Hi Santiago

not an easy question since normally we're interested in cancer genes.

for that purpose, I would go to the COSMIC database (https://cancer.sanger.ac.uk/cosmic), download the whole data and compare with a gene list obtained in the Ensembl website (http://www.ensembl.org/downloads.html). subtraction will give you what you need.

fred

The RUVSeq by Davide Risso (https://bioconductor.org/packages/release/bioc/vignettes/RUVSeq/inst/doc/RUVSeq.pdf) provide some interesting solutions to this problem. The relevance to your questions is that

• The package provide a way to empirically define the set of unchanged genes in the dataset itself and use them as a control.
• Alternatively, the package provide two other ways of calculating a normalization factor (not a list of genes) using replicates or residuals.

The first answer while technically correct is a bit misleading: COSMIC reports all somatic mutations found in cancer samples. This means that many of the genes touched by these mutations are not what people think when they speak of cancer genes: they generally want genes that are directly involved in driving a cancerous state (ie genes with driver mutations and not passenger mutations). So one solution is to substract from the complete list of human genes (on Ensembl, HGNC or NCBI) a master list of cancer genes. The problem is that there is not a single consensus list. COSMIC keeps a curated census list:

https://cancer.sanger.ac.uk/census

That contains 723 genes

But this is not complete and 2 years ago I tried to compile myself such a list and at that time I had already >3'000 genes in it.

Note that if you do go with the first answer your list will also be incorrect for the following reason: some genes are not 100% mapped correctly by Ensembl and thus Cosmic can not in some cases map somatic mutations to these genes. As there is a huge amount of data now in Cosmic almost all genes that are mapped have associated sequence variations, so the list you will obtain will contain a mixture of very small genes and incorrectly mapped genes.

Best

Amos

The answer above examplifies what I mean by saturation of "cancer" annotation: 17'910 protein coding genes associated with "cancer" that's 90% of all human protein coding genes. Thus this is not a list of cancer "genes" but a list of genes where eithet mutations where found in cancer samples or where transcription levels changed significantly in a cancer samples. Again this is true now of all human genes!