If you are looking for the normalised count matrix then please refer to this function in DESeq https://www.rdocumentation.org/packages/DESeq2/versions/1.12.3/topics/estimateSizeFactors

You can see an example here: https://github.com/uhkniazi/BRC_FOG1KO_John_PID_20/blob/921e1a3ea40061b8a2ecdad3332b431425c63330/09_exploratoryAnalysis.R#L76

Best wishes

You can use the function `count` with the argument `normalized = TRUE` to get the normalized counts matrix from the `DESeqDataSet` object. Check the section "Heatmap of the count matrix" in the package vignetter (http://bioconductor.org/packages/release/bioc/vignettes/DESeq2/inst/doc/DESeq2.html#quick-start)

Umar Niazi ,

I am an assistant professor in Ikuo’s department.

Thank you for your advice. However, we would like to show the heat maps of log fold-changes. So, we need to calculate log fold-change between controls and each experimental sample; This cannot be possible when normalized read count data of control is zero. The size factor seems to be used for data normalization which is calculated by dividing read count by the size factor. This method is good to draw a heat map of normalized read count data of each sample, showing control and experimental groups side by side. Are there any methods to make a heat map, using the log fold change data of each experimental sample, compared with controls, in DESeq2?

Mahmoud Ahmed ,

Thank you for your advice. We have read the package vignette of DESeq2 and understood a method to make a heat map, using read count data provided by DESeq2. According to the package vignette, an estimator of log fold-change, such as variance stabilizing transformation (vst) and regularized log transformation (rlog), seems to be used to make a heat map of read count data. Thus, we used ‘rlog’ and were able to obtain normalized log fold-change data of each sample and draw a heat map. However, we found that there were differences between the data by DESeq2 (one data per group, which should be the mean of 5 samples) versus rlog (the mean data calculated using rlog-normalized each fold-change data of 5 samples). For example, using DESeq2, log fold-change of CCL8 is 9.18 (this should be mean of 5 samples). Using rlog, log fold-changes of 5 samples are 4.95, 7.13, 7.22, 5.78, and 6.47 whose mean is 6.31. In theory, the original DESeq2 and rlog data should be the same data? Is it OK, using rlog?

The short version of this can be something like this:

> con = 0+1

> tr = 10+1

> log(tr)-log(con)

> log(tr/(con))

add a constant term to avoid divisions by zero.

The long version is to extract the log coefficients for the genes from the model object itself. For plotting purposes the short version should be sufficient if you add a constant number before taking log of the normalised matrix.