Hello, I am performing WGCNA on RNA sequencing data for a cohort. We are struggling with the normalization method that will not interfere with the correlation with the traits.

1. Which would be the most appropriate method to normalize RNA seq for WGCNA? I have come across where vst and rlog are being used. However, there was little explanation for

2. Why DESeq2 median normalization or quantile normalization was not used. We also have 4 separate batches of sequencing for which I am using RemoveBatchEffect after normalization.

3. In case if vst/rlog, shall vst or rlog be performed on matrix or deseq2 object?

Following are the codes for multiple ways that are used. Could you please correct me if it's wrong and comment on the right one to choose?

METHOD 1: USING CPM (library:edgeR)

data_filt <- cpm(data_filt, log=TRUE)

data_adj1 <- removeBatchEffect(data_filt, Batch = coldata$Batch, covariates=NULL[,-1])

----------------------------------------------------------------------

METHOD 2: USING QUANTILE NORMALIZATION (library:preprocessCore)

QST<- preprocessCore::normalize.quantiles(data_filt, copy = TRUE, keep.names = TRUE)

rownames( QST) <- rownames(data_filt)

colnames( QST) <- colnames(data_filt)

data_adj <- removeBatchEffect(QST, Batch = coldata$Batch, covariates=NULL[,-1])

------------------------------------------------------------------------

METHOD 3: USING DESEQ2 (library:DESeq2)

dds <- DESeqDataSetFromMatrix(countData = data_filt, colData = coldata, design = ~1) #~1 because i have a cohort and not case control

dds <- DESeq(dds)

norm_count <- counts(dds, normalized = TRUE)

data_adj <- removeBatchEffect(norm_count, batch = coldata$Batch, covariates=NULL[,-1])

------------------------------------------------------------------------

METHOD 4: USING VST (library:DESeq2)

dds <- DESeqDataSetFromMatrix(countData = data_filt, colData = coldata, design = ~1)

dds <- DESeq(dds)

vsd <-varianceStabilizingTransformation(dds) #requires matrix or DESeq object

vsd<-getVarianceStabilizedData(vsd)

data_adj1 <- removeBatchEffect(vsd, batch = coldata$Batch, covariates=NULL[,-1])

-------------------------------------------------------------------------

METHOD 5: USING RLOG (library: DESeq2)

rlog <- rlogTransformation(data_filt, blind = FALSE)

data_adj<- removeBatchEffect(rlog, batch = coldata$Batch, covariates = NULL[,-1])

--------------------------------------------------------------------------

Your help will be really appreciated in this matter, Thank you, Sukeshini K

WGCNA is still not part of Bioconductor (Batch adjustment for cohort based RNA seq data) so please check its documentation what the authors recommend. Generally, note that batch regression should be done on log2-scale data, so method 3 is wrong per se. WGCNA documentation is currently at https://bioinformatics.stackexchange.com/questions/21885/where-to-access-the-wgcna-tutorial-documents-horvath-lab-site-down/21886#21886