You can calculate the statistical power of your study based on the number of samples you have. I'm assuming that you have 85 adult medulloblastoma samples as there is already considerable literature on the consensus subgroups of paediatric medulloblastoma (SHH, WNT, Group 3 and 4), in which case you can just use their 'signature' to subgroup your patients, or download the publicly available data sets to compliment your samples (as a validation set).

If you do unsupervised sub-grouping it will also depend on the number of observations you use to classify your samples as you will have to correct for false discovery rate when the Affy chip you use has 10's of thousands of probes (for expression arrays), or a million probes (for SNP arrays).

There are lots of tools available to determine the robustness of your classification method such as non-negative matrix factorization, Kmeas, SOM, silhouette clustering...rand index (for comparing two clusters from different platforms). The list goes on...

Personally, I would not start of with expression arrays especially if you plan to add more samples to your analysis in the future. It is much more tedious to normalize and control for batch effect with expression arrays than with other arrays. Also, they are not as reproducible as one would like. My suggestion is (depending on your funding) is to do Illumina 450K methylation array first, Affymetrix SNP6.0 second and then an expression array (Illumina HT-12 or Affy).

I hope this helps. Best of luck.

Thank you very much!

But I don't have any money till now. Which means I can do nothing. I am trying to find some money,hehe