DESeq NB test for samples without biological replicates

[jason_ARGONAUTE]

Hi guys,

I hv encountered problems performing negative binomial test for samples without replicates while control group has replicates with DESeq.

Experimental design:
conduct the experiment under three conditions with common control: A B C (same samples treated at 3 distinct levels -- 50%, 80%, 95%)
Condition A has two biological replicates, Condition B and C have no biological replicates, Control group has two biological replicates.
noted as A_1, A_2, B, C, Ctrl_1, Ctrl_2
Pipeline of transcriptomic RNA-seq(paired-end):
map clean reads to ref genome using Tophat -> get the raw counts using SAMtools and HTSeq -> normalized the raw counts matrix using DESeq -> negative binomial test using DESeq

# Do stats based on a negative binomial distribution from Simon's vignette
With replicates:
cds = estimateDispersions(cds)
results = nbinomTest(cds, "condition", "control")
Without replicates:
cds = estimateDispersions(cds, method = "blind")
results = nbinomTest(cds, "condition", "control")

So, my confusion goes for a reasonable strategy to conduct NB test without replicates:
I have two replicates of control group(Ctrl1,Ctrl2) while no replicates for conditon B and conditon C, which way to perform a negative binomial test fits best for this case?
My current strategy is performing NB test between Ctrl1 and B/C(ignore Ctrl2).

[dpryan]

You'll be better off using the control replicates than if you were to use the "blind" method.