Question

DESeq2 contrast vs no contrast

0

Entering edit mode

ribioinfo ▴ 100

@ribioinfo-9434

Last seen 3.7 years ago

Hi I have a question about the use of the contrast of DESeq2.

If I have 3 (or more) conditions: A, B and C and I want to do: A vs C and B vs C

I can create a table with the columns A B C and to use the contrast on it. With this approach i can compare the FC of the two comparisons

If I made the tables A B and B C and I wanted to do: A vs C and B vs C now is it wrong to compare the FC of the two a comparisons?

Thank you.

deseq2 • 1.6k views

ADD COMMENT • link 8.3 years ago ribioinfo ▴ 100

score 0 · Answer 1 · 2016-01-12

0

Entering edit mode

Michael Love 41k

@mikelove

Last seen 6 hours ago

United States

You can compare fold changes across different contrasts, across different experiments, and so on. Fold changes are the estimates for how much the gene expression changes multiplicatively across the groups you compare.

ADD COMMENT • link 8.3 years ago Michael Love 41k

score 0 · Answer 2 · 2016-01-12

0

Entering edit mode

ribioinfo ▴ 100

@ribioinfo-9434

Last seen 3.7 years ago

Ok but in the first case i have one normalized table in the second i have two normalized tables. After the normalization the FC of the gene x in A vs C, using contrast on the table with the columns A B C, is different compared to the FC of the gene x calculated on the table with the columns A C without the column B.

So is it better to make one table and normalize all together or to make a table for every comparisons?

ADD COMMENT • link 8.3 years ago ribioinfo ▴ 100

0

Entering edit mode

Generally we recommend putting all the samples from all the different conditions into one dds object, then running DESeq(), then making comparisons with results(). The estimation of dispersion can often be better when you create one big dds object.

ADD REPLY • link 8.3 years ago Michael Love 41k

0

Entering edit mode

Thank you. I have another question. If my raw data table contains very low value, for example 1 reads only in one sample, is it better to filter the table before the normalization or such values are important for the statistical model?

ADD REPLY • link 8.3 years ago ribioinfo ▴ 100

0

Entering edit mode

Filtering if performed automatically in DESeq2 (see vignette) Basically the only reason to pre-filter with DESeq2 is to increase speed if you have hundreds of samples for example.

ADD REPLY • link 8.3 years ago Michael Love 41k