Dear Dr. Stark,

The following question has been asked before regarding error in dba.analyze step of diffBind when the contrast is set up for samples with no replicates. Obviously I should be comparing samples with multiple replicates but while I wait for more data from the sequencer it will be great if I could implement DiffBind for no replicates.

When I run dba.analyze step, I get the following error:

> dba.analyze(tf.chip)

Error in d$span : $ operator is invalid for atomic vectors
In addition: Warning messages:
1: Some groups have no replicates. Results may be unreliable.
2: In estimateCommonDisp(res) :
  There is no replication, setting dispersion to NA.
3: In estimateGLMCommonDisp.default(y = y$counts, design = design,  :
  No residual df: setting dispersion to NA
4: In estimateGLMTagwiseDisp.default(y = y$counts, design = design,  :
  No residual df: setting dispersion to NA
 

The warnings make complete sense but I am unable to figure out why it throws this error. I know you hate to deal with errors on data with no replicates but it will be great if you already know the solution for this one. See below for output of SessionInfo() command. Please let me know if more information is needed. Apologies for the trouble.

Cheers,

Danish.

> sessionInfo()

R version 3.1.0 (2014-04-10)
Platform: x86_64-unknown-linux-gnu (64-bit)

locale:
 [1] LC_CTYPE=en_GB.UTF-8       LC_NUMERIC=C              
 [3] LC_TIME=en_GB.UTF-8        LC_COLLATE=en_GB.UTF-8    
 [5] LC_MONETARY=en_GB.UTF-8    LC_MESSAGES=en_GB.UTF-8   
 [7] LC_PAPER=en_GB.UTF-8       LC_NAME=C                 
 [9] LC_ADDRESS=C               LC_TELEPHONE=C            
[11] LC_MEASUREMENT=en_GB.UTF-8 LC_IDENTIFICATION=C       

attached base packages:
[1] stats4    parallel  stats     graphics  grDevices utils     datasets
[8] methods   base     

other attached packages:
 [1] DiffBind_1.12.3         GenomicAlignments_1.2.2 Rsamtools_1.18.3       
 [4] Biostrings_2.34.1       XVector_0.6.0           limma_3.22.7           
 [7] GenomicRanges_1.18.4    GenomeInfoDb_1.2.4      IRanges_2.0.1          
[10] S4Vectors_0.4.0         BiocGenerics_0.12.1     BiocInstaller_1.16.2   

loaded via a namespace (and not attached):
 [1] amap_0.8-14        base64enc_0.1-2    BatchJobs_1.6      BBmisc_1.9        
 [5] BiocParallel_1.0.3 bitops_1.0-6       brew_1.0-6         caTools_1.17.1    
 [9] checkmate_1.5.2    codetools_0.2-11   DBI_0.3.1          digest_0.6.8      
[13] edgeR_3.8.6        fail_1.2           foreach_1.4.2      gdata_2.13.3      
[17] gplots_2.16.0      grid_3.1.0         gtools_3.4.1       iterators_1.0.7   
[21] KernSmooth_2.23-14 lattice_0.20-30    RColorBrewer_1.1-2 RSQLite_1.0.0     
[25] sendmailR_1.2-1    stringr_0.6.2      tools_3.1.0        zlibbioc_1.12.0  

I've had a look, the issue is in edgeR -- it can;t compute a dispersion estimate without replicates and a later call fails.

You can run this with DESeq2 [dba.analyze(th.chip,method=DBA_DESEQ2)] and it computes a result, but the adjusted p-values are all pretty high. You can use this to calculate fold changes and rank the sites if you wish, but without replicates we can't get meaningful confidence statistics.

Cheers-

Rory

Hi Danish-

I agree it shouldn't throw an exception. Could you send me a copy of your DBA object tf.chip so I can debug this?

Cheers-

Rory

Hi Rory,

Many thanks for the reply. I have emailed you .Rdata of tf.chip. Please let me know if more information is needed.

Cheers,

Danish.