Survival analysis based on a set of genes for TCGA data

Question

[fifty_fifty]

I need to analyze the survival of LUAD TCGA patients depending on the level of expression of a gene set (signature?). I have more than 30 genes for that.

I found this Survival analysis of TCGA patients integrating gene expression (RNASeq) data tutorial. But that is for one gene and KMplot is based on gene alteration rather than low vs high expression. I also used cBioPortal to plot survival for each gene.

I suspect that I need to look at my question as an ANOVA problem. Some genes will add significance to the set and some will not.

So, in the end, I would like to have a set of genes which combination will significantly change the survival of LUAD patients. Are there any readily available maybe web-based tools for that? if not, I will appreciate any advice on how to structure algorithm for my problemband using existing R packages for survival analysis or basically anything

Answer 1

[Hamid Ghaedi]

You may need first to look at your gene one by one to know whether they are significantly associated with survival probability or not (univariate analysis). Then picking significant genes and make a model with those genes (multivariate Cox regression analysis).

For univariate analysis check this RNA seq survival analysis in R. At the very end of the page, you may find a function that takes a list of gene and return survival analysis result as a table for you. I used this for more than 10,000 genes.

For multivariate analysis check this one.

The point would be how you dichotomize the expression data (usually people do this by considering z score). Once you got dichotomized data, the rest would be the same as other kinds of survival analysis.

[fifty_fifty]

thank you! I did survival analysis for each gene, but only couple of them are significant. So I thought maybe the combination of more genes will be more significant for survival.

[yussufhajjaj93]

Hi Hamid, thank you so much for the tutorial! But I wanted to ask if there a simple way to classify the data into quartiles or tertiles using the calculated Z-scores? Thanks in advance.