What SAS procedure can be used to analyze associations between microRNA (miRNA) expression profiles and a continuous variable (e.g., tumor size or certain scoring scale). I am trying to use SAS to find an association between a continuous variable (e.g., tumor volume) and miRNA expression profiles (about 200 miRNAs) obtained from almost 30 subjects' blood samples using qPCR. I already tried using Genesifter.net but it seemed to work only with categorical variables. Additionally, most of the articles I find related to miRNA expression profiles seem to compare categorical variables (e.g., treatment/no treatment) while I am interested into continuous ones.
What does your mRNA data look like? It's been a while since I took genetics.
I'm going to pretend like I know what the data looks like (scary thought;-) and assume that you have data showing that a certain variant of a gene is present. To simplify this assumption, I will assume that you have your data as Normal=0, Mutant = 1. If this is the case, you can try using Multiple Linear Regression. I would probably add how long the person has had their tumor to your analysis too. This could be valuable information unless tumors spontaneously generate to full volume after a short period of time.
If you only have 30 subjects to work with, you can only really look a few different RNA at a time. I would limit my analysis to 5-6 genes, unless you get more information(subjects). If you try using a series of correlations, 10 or so will be wrong because you have so many RNA to test. You will also find that certain RNA are correlated to other RNA. This makes testing multiple correlations among hundreds of RNA troublesome.
Dear Safa
I agree with Andrew. I see no problem in using correlation analysis between variables (miRNA concentrations and quantitative measurements of other variables).
I also agree about the number of samples vs number of variables. Perhaps you must compare far endpoints of your patients and determine which bunch of miRNA is differentially expressed, and then use the correlation analysis.
Best of luck.
Paco
Thank you for the rapid replies Mr. Ekstrom and Paco. The data is normalized data of qPCR replication cycles (delta CT) until the miRNA was detected. So I have 200 miRNAs, each one of them has a log transformed intensity in each of the 30 rats. My aim is to come up with the "5 to 6" miRNA that are significantly more or less expressed in subjects with a larger tumor volume (measured post-mortum) and other prognostic factors.
Hi Safa
It looks good... Try to get the differentially expressed miRNAs using composite extreme endpoints of your variable (tumour volume), and then analyze the variation of each of the selected miRNAs along tumor size...
Best of luck.
Paco
Safa,
If you have samples from other patients, you can use cluster analysis. The more closely two samples resemble each other, the tighter the cluster will be. You can use k-means clustering and set k=2, diseased and disease free. This would be another way to see if similar mRNA profiles correlate to disease.
- Badges
- Science topic
- Similar topics
- Engineering
- Modeling