# How to correlate survival probability to gene expression profile?

I am reading up survival analysis of cancer patients and correlating them to gene expression. I came across this tutorial. Its mostly code. So what I still have not been able to understand is how exactly is the clinical data and the gene expression data merged?

For example, I have a matrix of clinical data (m by n) where there are m patients and n columns related to followup days, death days etc. I also have an gene expression data (m by t) where m is the number of patients and t is the genes. I first find the survival probability using Kaplan Meier method and plot it. Now how do I related the survival probability to the expression data?

• As described in the link, you classify your samples into altered/not altered or high/low expressing (for your gene) by using Z scores (e.g. +/- 1.96) or a median cut-off, respectively. Now that you have a column that gives the expression status of the gene of interest, you can plot the survival for both conditions. The R code is in your link, and you can find how to do the same analysis with Graphpad Prism here: graphpad.com/support/faqid/1747 – srao Oct 27 '17 at 6:53
• @srao so you mean say out of 500 patients 200 patients will have high expression and 300 will have low, so the survival probability for high expression would involve only the patients having high expression and not those patient having low expression? – girl101 Oct 27 '17 at 8:03
• Yes, survival probability will be calculated separately for each group, that's how I understand it. In a more complicated example, you could have 4 groups (say 1 group for each quartile of gene expression). There are quite a few papers out there that have done this with TCGA data. – srao Oct 27 '17 at 10:57
• @srao true but how exactly it is done is not mentioned. papers mainly concentrate on results and what is done. detailed description is needed for first timers... – girl101 Oct 28 '17 at 6:11