**30**wrote:

Hi

I have the following experimental set up. I have RNA-seq time series data that span (0h,2h,....22h) and am interested in temporal patterns of gene regulation. However, the data was constructed by sampling repeatedly two cultures that are (in time) 11h apart. In other words, samples 0,2,..10 and 12,14,...,22 come from two different cultures but maintained nearly identically.

Since I have a multi-level design, I decided to capture the correlation between the two sets of samples. In addition, I use quality weights and based on previous BioC posts, I iterated the duplicateCorrelation and `voomWithQualityWeights`

twice.

I have the two following questions:

1. Is it reasonable to do both duplicateCorrelation and quality weights?

2. Should I use the same blocking I use for duplicateCorrelation also with Quality weights?

3. Can the `consensus.correlation `

be negative? Depending on the design matrix I used I get either positive or negative correlation values` and `

the latter gives an error in` chol.default(V). `

What is recommended in this case?

4. I tried to test if the FDR control are reasonable, by shuffling the time labels and estimating the the number of false positives and I realised that actual FDR was much higher than that designed for?

Thank you in advance!

**My code snippet:**

# count matrix has dimension 9000 x 12 T <- 22 time <- seq(0,T,by=2) block <- rep(c(1,2), each = 6) in.phase <- cos(2*pi/T*time) out.phase <- sin(2*pi/T*time) y_harm <- calcNormFactors(y_harm) design <- model.matrix(~in.phase + out.phase) v_harm <- voomWithQualityWeights(y_harm, design) corfit <- duplicateCorrelation(v_harm, design, block = block) v_harm <- voomWithQualityWeights(y_harm, design, block = block, correlation = corfit$consensus.correlation) corfit <- duplicateCorrelation(v_harm, design, block = block) fit_harm <- lmFit(v_harm, design, correlation = corfit$consensus.correlation, block = block) fit_harm <- eBayes(fit_harm, robust = robust, trend=TRUE)

**21k**• written 4 months ago by BharathAnanth •

**30**