Question: deseq2 outliers problems
gravatar for aristotele_m
2.5 years ago by
aristotele_m30 wrote:

Dear all,

I have this situation for a gene overexppressed:

Group A: average 85.33

GroupB: average 23081.19

average gene 1930.54

On the result table  of differential expression:

basemean: 1930.54,log2Foldchange 5.115 lfcSE 0.341

The results seem more different from  the ratio obtained.. 

I found this results also in other comparison:

89.267 21448.225


baseMean log2FoldChange lfcSE stat pvalue padj
1794.7910 1.70 0.175111091 9.7157783091 2.58268280887747E-22 3.97733152567131E-18

Any idea?




R version 3.3.2 (2016-10-31)
Platform: x86_64-pc-linux-gnu (64-bit)
Running under: Ubuntu 16.04.2 LTS

 [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C               LC_TIME=it_IT.UTF-8       
 [7] LC_PAPER=it_IT.UTF-8       LC_NAME=C                  LC_ADDRESS=C              

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

other attached packages:
 [1] gplots_3.0.1               genefilter_1.54.2          limma_3.28.21             
 [4] biomaRt_2.28.0             reshape2_1.4.2             RColorBrewer_1.1-2        
 [7] ggplot2_2.2.1              pheatmap_1.0.8             DESeq2_1.12.4             
[10] SummarizedExperiment_1.2.3 Biobase_2.32.0             GenomicRanges_1.24.3      
[13] GenomeInfoDb_1.8.7         IRanges_2.6.1              S4Vectors_0.10.3          
[16] BiocGenerics_0.18.0       

loaded via a namespace (and not attached):
 [1] gtools_3.5.0         locfit_1.5-9.1       splines_3.3.2        lattice_0.20-35     
 [5] colorspace_1.3-2     htmltools_0.3.5      base64enc_0.1-3      survival_2.40-1     
 [9] XML_3.98-1.5         foreign_0.8-67       DBI_0.6-1            BiocParallel_1.6.6  
[13] plyr_1.8.4           stringr_1.2.0        zlibbioc_1.18.0      munsell_0.4.3       
[17] gtable_0.2.0         caTools_1.17.1       htmlwidgets_0.8      memoise_1.0.0       
[21] labeling_0.3         latticeExtra_0.6-28  knitr_1.15.1         geneplotter_1.50.0  
[25] AnnotationDbi_1.34.4 htmlTable_1.9        Rcpp_0.12.9          KernSmooth_2.23-15  
[29] acepack_1.4.1        xtable_1.8-2         backports_1.0.5      scales_0.4.1        
[33] checkmate_1.8.2      gdata_2.17.0         Hmisc_4.0-2          annotate_1.50.1     
[37] XVector_0.12.1       gridExtra_2.2.1      digest_0.6.12        stringi_1.1.2       
[41] grid_3.3.2           tools_3.3.2          bitops_1.0-6         magrittr_1.5        
[45] lazyeval_0.2.0       RCurl_1.95-4.8       tibble_1.2           RSQLite_1.1-2       
[49] Formula_1.2-1        cluster_2.0.6        Matrix_1.2-8         data.table_1.10.0   
[53] assertthat_0.1       rpart_4.1-10         nnet_7.3-12        
deseq2 • 484 views
ADD COMMENTlink modified 2.5 years ago • written 2.5 years ago by aristotele_m30

Can you give us more details to understand the problem you are facing?

ADD REPLYlink written 2.5 years ago by Bio_Ram0

Gruop  A  express GFP and Group B express my target genes. So I want to understand the effect of  overexpression of my gene on  my transcriptome.

I know are overexpressed and also my count demonstate are overexpressed but the fold change are not close with my ratio calclulate from the counts. Is it normal?

ADD REPLYlink written 2.5 years ago by aristotele_m30
Answer: C: deseq2 outliers problems
gravatar for Michael Love
2.5 years ago by
Michael Love26k
United States
Michael Love26k wrote:

If you used the latest version 1.16 you would get an MLE fold change. You can set betaPrior=FALSE to get an MLE fold change with older versions.

ADD COMMENTlink written 2.5 years ago by Michael Love26k

Thanks now works. But I have only this series where I need that comand. What could be the reason? Is a bug or  don't change so much because I have always I very  low p-value (^-18)


ADD REPLYlink written 2.5 years ago by aristotele_m30

It's not a bug it's a feature. Likely the MLE fold change is driven by an outlier. We provide moderated LFC estimates (previously using DESeq(), now using lfcShrink() after DESeq()). We can show that moderation of LFC produces more reproducible estimates:

ADD REPLYlink written 2.5 years ago by Michael Love26k
Please log in to add an answer.


Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 16.09
Traffic: 225 users visited in the last hour