Hi,

I have RNA-seq data with three factors

• line (genotype): a554, s018, s311, s84
• tissue : leaf, sam
• condition : cold, regrowth

Thanks to DESeq documentation and r-bloggers page I have quite good understanding of model for two-way analysis (ie. two factors and interaction).

However for three factors situation complicates a lot.

We are primarily interested in a question: does lines (genotypes) react differently to condition? But also: does line reaction to condition is different in tissues? And also in simple cold vs regrowth comparison for each line. So I should investigate the interaction, it is not hard for two-factor analysis but for three-way it is above my skills...

So I wanted to use method proposed in DESeq2 section on contrasts

Where DESeq2 developers combine factors into one factor 'group'. Indeed it makes pairwise contrasts very convenient.

So here is an extract from my colData

colData(dds2d.ch.rg.group)
DataFrame with 48 rows and 11 columns
          pool.plant      letter   tissue     line   pool.time     time
         <character> <character> <factor> <factor> <character> <factor>
la5.3.21     a5.3.21           C     leaf     a554       10-17        3
ls0.3.21     s0.3.21           F     leaf     s018       10-17        3
ls8.3.21     s8.3.21           I     leaf     s84        10-17        3
ls3.3.21     s3.3.21           L     leaf     s311       10-17        3
sa5.3.21     a5.3.21           O     sam      a554       10-17        3
...              ...         ...      ...      ...         ...      ...
ls3.3.43     s3.3.43           L     leaf     s311       10-17        3
sa5.3.43     a5.3.43           O     sam      a554       10-17        3
ss0.3.43     s0.3.43           S     sam      s018       10-17        3
ss8.3.43     s8.3.43           W     sam      s84        10-17        3
ss3.3.43     s3.3.43           Z     sam      s311       10-17        3
              day      rep condition              group sizeFactor
         <factor> <factor>  <factor>           <factor>  <numeric>
la5.3.21        2        1      cold     a554.cold.leaf   0.599133
ls0.3.21        2        1      cold     s018.cold.leaf   0.532561
ls8.3.21        2        1      cold     s84.cold.leaf    0.440120
ls3.3.21        2        1      cold     s311.cold.leaf   0.569728
sa5.3.21        2        1      cold     a554.cold.sam    1.329649
...           ...      ...       ...                ...        ...
ls3.3.43        4        3  regrowth s311.regrowth.leaf    1.12049
sa5.3.43        4        3  regrowth a554.regrowth.sam     1.82546
ss0.3.43        4        3  regrowth s018.regrowth.sam     2.33982
ss8.3.43        4        3  regrowth s84.regrowth.sam      2.14273
ss3.3.43        4        3  regrowth s311.regrowth.sam     1.96530

My design is design(dds2d.ch.rg.group)= ~ 0 + group

And my coefficients are

dds2d.ch.rg.group=DESeq(dds2d.ch.rg.group)
resultsNames(dds2d.ch.rg.group)
 [1] "groupa554.cold.leaf"     "groupa554.cold.sam"     
 [3] "groupa554.regrowth.leaf" "groupa554.regrowth.sam" 
 [5] "groups018.cold.leaf"     "groups018.cold.sam"     
 [7] "groups018.regrowth.leaf" "groups018.regrowth.sam" 
 [9] "groups311.cold.leaf"     "groups311.cold.sam"     
[11] "groups311.regrowth.leaf" "groups311.regrowth.sam" 
[13] "groups84.cold.leaf"      "groups84.cold.sam"      
[15] "groups84.regrowth.leaf"  "groups84.regrowth.sam"

So I understand that for comparison groupa554.cold.leaf vs groupa554.regrowth.leaf i should use command

results(dds2d.ch.rg.group, contrast=c('group', 'groupa554.cold.leaf', 'groupa554.regrowth.leaf'))

Is it possible to get interaction with such design?

So for example could I compare if a554 reaction to regrowth in leaf is different to s018 reaction like that

(groupa554.regrowth.leaf - groupa554.cold.leaf) - (groups018.regrowth.leaf - groups018.cold.leaf)

Or to find out if a554 reaction to regrowth is different in leaf and sam

(groupa554.regrowth.leaf - groupa554.cold.leaf) - (groupa554.regrowth.sam - groupa554.cold.sam)