Hi.

I have a comparison between two groups of samples that after an initial DE analysis, show a ton of differentially expressed genes. After checking with Quantro, I observe that the distribution of gene expression is too different so, according to Quantro conclusions, general adjustment methods (like quantile, Variance-stabilizing transformation, etc) should not be used.

I in this situation, I wonder if using DESeq2 is correct and possible (using default configuration) or if it's better to adjust it somehow. According to DESeq2 and DESeq papers, the size factors calculation with the median of ratios solves the problem of having "a few highly and differentially expressed genes that may have strong influence on the total read count" but what happens when the overall distribution of expression for the two groups is so different. Should the size factors be adjusted by other methods?

Thank you very much in advance.

If you think there is a global change in expression across condition, you’d need to have some genes that you can identify as not changing, or use spike ins for normalization. Do you have these? They can be provided as ‘controlGenes’.

On a more philosophical note, the question "which genes are differentially expressed?" that methods like DESeq2 are trying to answer, is scientifically most meaningful if the answer is relatively sparse, and identifying such genes bears some degree of information content. If almost everything is differentially expressed, then a more global, or entirely different, question might be scientifically more meaningful.

Thank you very much for all the answers.

Unfortunately, we don't have spike-ins but we'll continue working on it.