[BioC] Unusual results with DESeq ?

Simon Anders anders at embl.de
Fri Sep 13 11:32:09 CEST 2013


Hi Osvaldo

Looking through the excerpt of your result list, it seems that your read 
counts are very low. Most miRNAs seem to have less than 50 reads, so it 
might well that you simply have not sequenced deeply enough.

Have you enriched your library for short RNA molecules, or is this 
standard RNA-Seq? For how many miRNA genes do you have counts, and how 
many are above 50?

For now, independent filtering might help: Exclude all the genes with 
less than, say, 20 or 40 counts (according to baseMean), and run the 
Benjamini-Hochberg adjustment only on the remaining ones. I think we 
have a chapter on the vignette about this.

Also double-check teh normalization with MA plots. With so small counts, 
the default location measure (the median) might work subobtimal; maybe 
try the shorth. (See ?estimateSizeFactors).

Also consider switching to DESeq2, which has better inferential power 
due to an improved dispersion estimation scheme.

   Simon


On 13/09/13 10:15, Osvaldo Graña wrote:
> hi there !!
> I am analyzing miRNA-seq data with DESeq, and I am getting no significant
> results, and I cannot see why is so.
>
> My experiment has two conditions and two replicates per condition:
>
> I execute it in R as follows:
> *countTable = read.table (
> "/mnt/TB3/ograna/Ozge_Uluckan/miRNA-seq/Analysis/tables_for_DESeq/cOB1_vs_tOB100"
> , header=TRUE, row.names=1)
> experiment_design = data.frame(
> row.names = colnames(countTable),
> condition=c("OB1","OB1","OB100","OB100"),
> libType=c("single-end","single-end","single-end","single-end")
> )
> library("DESeq")
> condition=factor(c("OB1","OB1","OB100","OB100"))
> cds = newCountDataSet( countTable, condition )
> cds = estimateSizeFactors( cds )
> normalizedReadCounts = counts( cds, normalized=TRUE )
> write.csv( normalizedReadCounts,
> file="cOB1_vs_tOB100.DESeq_normalizedReadCounts.csv" )
> cds = estimateDispersions( cds )
> res = nbinomTest( cds, "OB1", "OB100" )
> write.csv( res, file="cOB1_vs_tOB100.DESeq_diffExp.csv" )*
>
>
> I am getting the following size factors:
>> sizeFactors(cds)
>     OB.1OU    OB.2OU OB100.1OU OB100.2OU
>          1         1         1         1
>
>
> the differential expression table sorted by 'padj' is as follows (I am
> showing just a small set):
>
> number    id    baseMean    baseMeanA    baseMeanB    foldChange
> log2FoldChange    pval    padj
> *340    mmu-miR-204-5p    31.5    12.5    50.5    4.04    2.014355293
> 6.466284630369E-005    0.122859408
> 69    mmu-miR-1247-5p    57.5    32.5    82.5    2.5384615385
> 1.3439544012    0.0010710548    1
> 170    mmu-miR-15b-3p    99.25    135.5    63    0.4649446494
> -1.1048691179    0.0024355731    1
> 363    mmu-miR-214-3p    591.25    767    415.5    0.5417209909
> -0.8843781007    0.0040302141    1
> 1122    mmu-miR-664-3p    55.5    76    35    0.4605263158
> -1.1186444965    0.0065510324    1
> 630    mmu-miR-335-3p    9.5    15.5    3.5    0.2258064516
> -2.1468413883    0.0093262341    1
> 997    mmu-miR-574-3p    282    364.5    199.5    0.5473251029
> -0.8695300681    0.0101138488    1
> 176    mmu-miR-17-3p    29    41    17    0.4146341463    -1.2700891634
> 0.0109905451    1
> 1898    mmu-miR-99a-5p    1254.5    1568.5    940.5    0.5996174689
> -0.7378856803    0.0140500521    1
> 846    mmu-miR-467a-5p    11    17    5    0.2941176471    -1.7655347464
> 0.019028208    1
> 1900    mmu-miR-99b-5p    6246    7702.5    4789.5    0.6218111003
> -0.6854517236    0.0204136071    1*
>
>
> There is no even one significant miRNA. Is it that I am missing something
> or doing something wrong? What does it mean that all the 'padj' values are
> '1' with the exception of the first one '0.122' ?
> Is there a way to change the method used to correct the p-values?
>
> thanks very much in advance !!!
> regards.
>
>
>
> _______________________________________________
> Bioconductor mailing list
> Bioconductor at r-project.org
> https://stat.ethz.ch/mailman/listinfo/bioconductor
> Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor
>



More information about the Bioconductor mailing list