[BioC] Huge variation in results among contrasts - Please Help!

Gordon K Smyth smyth at wehi.EDU.AU
Sun Jun 9 04:25:26 CEST 2013

Dear Atul,

The description of your analysis doesn't make sense to me.  How could you 
have t-statistics for individual samples P0 and P30?

If you want help with code, then please give code.  See the posting guide.

The software developers only give advice on using code -- we don't analyse 
or trouble-shoot individual experiments on the list.

Best wishes

--------- original message ------------
Atul Kakrana atulkakrana at outlook.com
Thu Jun 6 05:42:15 CEST 2013

Hi Everybody,

I am facing problem in understanding the reason for difference among 
samples when sorted on basis of T-stats. When same samples are sorted on 
Fold Change than they show consensus in top genes.

There are total of 7 samples (Illumina microarray), 5 (P8,P12,P20,P42,P52) 
from one study and two (P0 and P30) from other two studies. So, 7 samples 
from 3 studies and one control. I performed standard micro-array analysis, 
imported samples ( 7 samples) from separate files (3 files), performed 
normalization, combined using 'lumi' package, filtered out absent probes, 
reduced probes to gene level and used 'limma' to perform decideTests.

As usual, I ranked genes on basis of T-stats. While sorting genes on basis 
of these ranks I found that if sorted on any of two columns (P0 and P30) 
there is a clear pattern of matching ranks between these two columns but 
totally different from other 5 columns. So, these 2 columns show consensus 
in ranks and other 5 columns show consensus in rank. Please see attached 
result snippet.

But when sorted by Fold change of any sample, all the samples show 
consensus by showing same top genes. It does not matter which sample is 
used for fold change based sorting the genes are (almost) same i.e 
consensus among samples. If you see that attached file which is sorted on 
fold change, you will notice a pattern in Ranks (Column C-G) among samples 
i.e. sample P0 and P30 showing similar ranks whereas all other showing 
different ranks.

So, my question is why samples when ranked with t-stats are not showing 
consensus where as when ranked with Fold Change shows consensus in top 
genes. Is it a batch effect?

I would really appreciate your help in understanding the problem.


The information in this email is confidential and intend...{{dropped:4}}

More information about the Bioconductor mailing list