0
$\begingroup$

After I got dds from my count matrix by DESeq(), I use results() method to get Padj and log2 fold changes for volcano plotting, then I found some gene's Padj values are infinite and the P-value is very close to 0; this makes those dots on my volcano plotting located very top to Y-aix and also confused me:

  1. is this normal or not, such a p-value (very close to 0) and -logPadj is infinite;
  2. how should I treat those data: are they very different expressed between samples and control or just abnormal data which should be taken off?

Here follows the table: enter image description here

table 1 is dds results, -log10(padj) show inf; table 2 is count matrix from featureCount table 3 is size factor of samples

Please help me !

$\endgroup$
6
  • $\begingroup$ It is not normal to get p-value close to zero, especially if you have not a lot of samples. If I look at your table 2, the mean counts don't tally with the genes in table 1... so can you sort this out $\endgroup$
    – StupidWolf
    Nov 27 '20 at 10:26
  • $\begingroup$ there are some instances, for example you knock out or over-express a gene, then of course you find this insanely differentially expressed.. $\endgroup$
    – StupidWolf
    Nov 27 '20 at 10:28
  • $\begingroup$ The SE are really low in this case though, rather than huge LFC. Very consistent gene expression? $\endgroup$
    – Greg
    Nov 27 '20 at 13:41
  • $\begingroup$ For log(0) error add an arbitrarily small number to the 0 p.adj values (e.g.10^-99) $\endgroup$
    – Greg
    Nov 27 '20 at 13:41
  • $\begingroup$ @StupidWolf Thank you StupidWolf and Greg for answer my question. I didn't list all the samples out, and one of the compounds has the opposite effect from the others, like: d has a positive effect on empty control, and a b c has a negative effect on empty control; According to the answers, I think this may cause the so much differentially expressed(but i'm not so sure). $\endgroup$
    – Chu
    Dec 4 '20 at 9:45
1
$\begingroup$
  1. The fold changes for those are huge (on the order of 32x for the largest), so it's unsurprising for the adjusted p-values to be absurdly small.
  2. There's no reason to treat them differently.
$\endgroup$
1
  • 1
    $\begingroup$ Agreed. If you want to get rid of Inf you can add a tiny number to the pvalues before the conversion, e.g. p + .Machine$double.xmin. $\endgroup$
    – ATpoint
    Nov 27 '20 at 17:34

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.