Number of driver mutation required to call a gene driver

Question

[Gene_MMP8]

I am currently working on an algorithm to distinguish driver from passenger mutations. I have a set of new genes that are like a validation set to me. After I have run the algorithm on these genes, I will have a set of predicted drivers and passengers. My question is, exactly when can I call a gene driver based on the number of driver and passenger mutation I get? Intuition tells me, I can call a gene driver even if it has a single driver mutation. Am I correct? Or is there any cutoff?

Answer 1

[Kevin Blighe]

You won't find any single answer to this. A gene driver, by definition, 'drives' / promotes tumour growth and / or progression. The term driver is used generally, though. A driver gene may have just a single point mutation, or may have no mutation at all and be regulated through other epigenetic means (or by a mutation in an intergenic regulatory region). So, going by the number of mutations is not the way to view this problem of driver gene identification.

Certain genes accumulate somatic mutations at higher frequencies than other genes, but these may have no role in tumour growth / progression.

For your algorithm, you will therefore have to include other features, such as epigenetic and other regulatory marks. Look up CADD in order to get a few ideas of what you could do.

Kevin

[Gene_MMP8]

Thanks for your answer. The problem is I am working with mice samples and just have the vcf data with me and have annotated driver/passenger mutations using SIFT. So its very hard to validate the results. So can you also suggest some ways to validate this?I have a set of predicted driver/passenger mutations for each gene in the test data.

[Kevin Blighe]

How have you annotated driver/passenger mutations using SIFT? - just going by the SIFT prediction? I am not sure that will be valid for any reputable journal, or even as a PhD thesis (if that is what you are doing). There are in silico prediction tools that are specifically tailoured for somatic variants. Take a look at the bottom, here: A: pathogenicity predictors of cancer mutations

Do you have a supervisor who is assisting you?

[Gene_MMP8]

Thanks again for your help. I am concluding a mutation to be driver if it is 'deleterious' according to SIFT or passenger if it is 'tolerated'. So are you suggesting a consensus of the predictions from all these tools will be more useful? This is a summer project and my supervisor is not there to help me. So I am going through papers and Biostars posts to get an idea.

[Kevin Blighe]

Okay, it is a summer project. Are you hoping to publish it or is it purely for training? You should at least try some of the other tools that I mentioned, if you can, in particular GWAVA and Funseq2. Using CADD is also generally good.

I think that it would be good to take a consensus, so, for example, choose 5 tools, and then require that a driver must have at least 3 of these predicting pathogenicity.

[Gene_MMP8]

Thanks a lot for your patience. I will definitely do a consensus and look for a more pronounced set of drivers.