Denosing Mirna Microarray Data From Blood Samples

Question

[khan]

Hi all,

I have miRNA microarray data from from blood sample. after normalization I saw quite different expression patterns for the same type of cancer among different patients(which I would expect to see at least some consistent expression pattern).

I want to find hidden pattern inside of this data set. what is the best approach to denoise and find the pattern? I applied different clustering and dimensionality reduction algorithms but I couldn't find robust pattern.

What are your suggestions?

Answer 1

[pld]

You could try t test each probe to see if in the tumor set that probe is significantly different than your healthy set. You might not see more than a few probes meet this level. You could use machine learning approaches to try and extract more complex situations but that might be a harder argument to support initially and validate later.

It is possible that there simply isn't a pattern at the level you are looking, or at all. Any number of differences between your patients could be causing significant differences in miRNA expression. On the other hand there may not be any differences. A lack of an obvious pattern doesn't mean your signal to noise ratio is poor, it could just mean that there isn't any signal to begin with. If your samples were as closely matched as possible and you still don't find anything, this suggests there isn't anything.

Answer 2

[QVINTVS_FABIVS_MAXIMVS]

Maybe you are being to liberal with plotting? Sometimes restricting your values and replotting them (heatmaps normalize by row) can bring out patterns. But it's important to note this alteration.