Hi, apologies in advance, I have read a bunch of the limma/dream tutorials on here, but none seem to exactly fit my question. I can adjust this to do duplicateCorrelation with limma if it seems more appropriate.

I have a repeated/paired measures design (Pre and Post an Intervention) and then a certain score (scaled here) where higher numbers are associated more severe disease. I want to model 2 things: 1) Genes that are up/down regulated with baseline severity. 2) Genes that are up/down regulated with change in severity from before to after the intervention

Based on the tutorials and all the posts here for limma/dream, rather than doing the Timepoint*Score Interaction, I tried to make it more interpretable this way:

#The Final Model here would contain the (1|Sub_ID) Random Effect and some other control vars. 
head(model.matrix(~0 + Timepoint + Timepoint:Score, data = metadata)) 

         TimepointPre TimepointPost           TimepointPre:Score        TimepointPost:Score
S1_1            1             0                   1.1148724                    0.0000000
S1_2            0             1                   0.0000000                    1.0429451
S2_1            1             0                   0.8990906                    0.0000000
S2_2            0             1                   0.0000000                    0.8990906
S3_1            1             0                   0.7552361                    0.0000000
S3_2            0             1                   0.0000000                    0.6473453


contrasts <- c(
      BL_SEVERITY = TimepointPre:Score,
      IMPROVEMENT = TimepointPost:Score  - TimepointPre:Score 
    )

I have 2 questions: 1) is this the appropriate way to determine my differential expression at baseline and pre-post intervention within the same model? 2) Given that "Improvement" is associated with a reduction in the score from pre to post, could I simply flip the contrast (i.e., IMPROVEMENT = TimepointPre:Score - TimepointPost:Score) so that the final DEGs could be interpreted as "genes which are upregulated with improvement for interpretability?