I understand that the presence of negative log-transformed count values is a common occurrence in RNA-seq data analysis and these values are not necessarily problematic for statistical analyses. I used
TMM method > log transformed and used limma for identifying significant gene expression (see below). Then, I extracted few genes for (line plot) that is high expressed between the groups of interest, but I see negative log (CPM) values on y-axis (though the expression pattern on the line plot shows the difference). My collaborators were not convinced with this y-axis scale and asked me if there is a way to get the values on the positive scale.
Then, I increased
prior.count (started increasing from +150 --> + 250), the minimum values increase to positive value scale. Meaning the log negative values now transformed to the positive values. However, I am not sure if adding larger prior.count value would affect downstream analysis like differential analysis in general. Does, choosing a large prior.count for visualization only seems like a good strategy because with prior.count = 1, the scale is negative.
Alternatively, I tried Deseq2 (VST) as it did not produce log negative counts. I also have TPM values and observed log of TPM also does not produce negative values. But I do not want to use these methods again for the analysis just to justify the visualization needs.
y <- calcNormFactors(y, method = "TMM") logCPM <- cpm(y, prior.count=1, log=TRUE) Use limma to analyse multi-factor and varaibles