ComBat â - multi-modal prior distribution

[minyoung lee [guest]]

Dear users

I have 3 batches and each batch have very different number of samples from 3 conditions. 

batch1=6 samples from conditionA and 8 samples from conditionC.
batch2=10 samples from conditionA and 3 samples from conditionB.
batch3=2 samples from conditionA and 2 samples from conditionB.

I just want to compare conditionA and conditionB. 
At first, I made mod and mod0 with all samples, so 3 batches and 2 categorical covariates were found. However, in this case ComBat did not work. So I remove conditionC in batch1. At this time, ComBat did work (Found 3 batches and Found 1  categorical covariate). But, the prior distribution is weird and when I did clustering and PCA analysis, the samples from conditionA in batch1 were very different from the others (all the others were mixed). In my case, the ComBat did not work? Is there anything I can do for correct results?

The ComBat result distribution figures are in the follwing link.
https://docs.google.com/drawings/d/1YKYfTwMOeoVONbbyd6BPKLEc5aBu1kpGRAUGJll6dwE/edit?usp=sharing





 -- output of sessionInfo(): 

R version 2.14.2 (2012-02-29)
Platform: i386-pc-mingw32/i386 (32-bit)

locale:
[1] LC_COLLATE=Korean_Korea.949  LC_CTYPE=Korean_Korea.949   
[3] LC_MONETARY=Korean_Korea.949 LC_NUMERIC=C                
[5] LC_TIME=Korean_Korea.949    

attached base packages:
[1] splines   stats     graphics  grDevices utils     datasets  methods  
[8] base     

other attached packages:
[1] bladderbatch_1.0.2 Biobase_2.14.0     limma_3.10.3      
[4] pamr_1.54          survival_2.37-4    cluster_1.14.3    
[7] sva_3.0.2          mgcv_1.7-22        corpcor_1.6.4     

loaded via a namespace (and not attached):
[1] grid_2.14.2     lattice_0.20-10 Matrix_1.0-5    nlme_3.1-108   


--
Sent via the guest posting facility at bioconductor.org.