Question

significant genes with DESeq2 plenty of outliers and zero counts

0

Entering edit mode

aec ▴ 90

@aec-9409

Last seen 4.5 years ago

Dear all,

I run DESeq2_1.18.1 with default options and obtained odd results. The significant transcripts (FDR<5%) are due to one or two outlier samples.

I checked the expression values and all samples except the outliers have 0 counts,

How can I deal with this situation? Which parameters should I modify? changing Cook's distance? disable independent filtering?

Thanks,

zero counts deseq2 odd results outliers • 1.6k views

ADD COMMENT • link updated 6.9 years ago by Michael Love 43k • written 6.9 years ago by aec ▴ 90

score 0 · Answer 1 · 2018-02-07

0

Entering edit mode

Michael Love 43k

@mikelove

Last seen 7 days ago

United States

It could be you have too few samples for the outlier detection to kick in, or the counts are not high enough to count as outliers. Either way, you can perform more aggressive filtering to remove genes with a count in only one or two samples:

dds <- estimateSizeFactors(dds)
keep <- rowSums(counts(dds,normalized=TRUE) >= 10) >= 3
dds <- dds[keep,]

ADD COMMENT • link 6.9 years ago Michael Love 43k

0

Entering edit mode

Hi Michael,

I have 20 samples, 10 per group ( I think DEseq2 requires minimum of 7 for outliers).

and the normalized counts of 2 significant genes:

	s1	s2	s3	s4	s5	s6	s7	s8	s9	s10	s11	s12	s13	s14	s15	s16	s17	s18	s19	s20
transcript1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	4706
transcript2	0	0	0	0	0	0	0	0	0	0	0	268.4	0	0	0	0	0	0	0	0

they appear to be high enough...

ADD REPLY • link 6.9 years ago aec ▴ 90

0

Entering edit mode

It's hard to say for a given gene why filtering/replacement does or does not apply, because it depends also on the dispersion of all the other genes. Nevertheless, the above code should help you here.

ADD REPLY • link 6.9 years ago Michael Love 43k

0

Entering edit mode

Yes, the above code helped me, but still, I do not understand why DESeq2 is not able to deal with these cases with default options.

ADD REPLY • link 6.9 years ago aec ▴ 90