I have RNAseq data from cells treated with 3 drugs , drugA, drugB, drugC and different combinations of those drugs so in total 8 arms, drugA, drugB, drugC, drugA+drugB, drugB+drugC, drugA+drugB+drugC, drugA+drugC and no treatment/control arms. I have 3-5 replicates per arm.

I am using DEseq2 for analysis but I am not sure what should be the design. Is a simple model with treatment as a single factor with 8 levels enough or a more complex model is more appropriate?.

Is LRT appropriate?. I tried to compare it to simple model using the code below and I found a lot more significantly expressed genes than using the simple model. Also on biological level, there mono therapy drugs are very different from the combinations with some resulting in additive effect and some combinations resulting in contradictory effect.

```
dds_LRT <- DESeq(ddsobj, test='LRT', reduced=~1)
res <- results(dds_LRT, lfcthreshold=0)
```

Can I obtain pair-wise comparisons from the dds_LRT to know the DEgenes are differentially expressed among which groups using the code below?.

```
res_LRT <- results(dds_LRT, contrast= list('drugA+drugB+drugC' , 'drugB+drugC'))
```

the aim of using the above code to study the effect of adding drugA to the combinations.

Is that correct?.

I am using DESeq2 version 1.32

Just to add

if dds_LRT is the ddsobject generated using LRT test and dds is the ddsobject generated by Wald test , there are a lot more Differentially expressed genes between the following 2 codes

```
res_LRT <- results(dds_LRT, contrast= c('Treatment', 'drugA+drugB+drugC' , 'drugB+drugC'))
res <- results(dds, contrast= c('Treatment', 'drugA+drugB+drugC' , 'drugB+drugC'))
```

Thanks

And further questions here:

DESeq2 likehood ratio