Hi all, I am sorry for the long post, but I am hopping i got all needed information and related questions and concerns. The main biological question is this, after a treatment is the ratio of H3K36me3 Vs polII changes and is the change significant? Which genes are the once mostly affected. I am providing a full That would be "easy" with a limma approach. Which I did, but the results either are random, so the treatment has random effect, (AND HOW CAN I TEST this ?) or due to the data are worthless.

But, I noticed that the distribution of OlII Vs H3K36me3 is not that linear and there are 2 different "clouds". Thinking of it I assume that is expected, between H3K36me3 Vs polII, since there are in principle genes with low transcription and higher. I am providing 2 graphs that I believe illustrate the story. The pvalue in the violin plot is from Wilcoxon test. This is my first question and concern. I am unsure with this kind of distribution if even that test is correct where number of events = around 40k. What do you all think, and what should I do for these kind of distribution? Should I separate the clouds and how to "properly" separate the 2 clouds?

I tried using kmeans with the following command, that seems to work. I am telling it to brake them in 3 groups, since there is a smaller one that is being identified as different.,:

km_clusters <- kmeans(log2(merged_df$Mean_H3_DMSO + epsilon), centers = 2)

I tested the hclust that also works, still not optimal. the DBSCAN method failed, OR I really do not know how to work with it properly.

I have normalised the counts using the EDASeq packaged by PolII and H3K36me3 separately, nothing really changed since the 2 clouds are a biology inherited thing.

  set <- EDASeq::withinLaneNormalization(set, "gc", which = "full")
  set <- EDASeq::withinLaneNormalization(set, "length", which = "full")

The final 3 plots are separation of the plots by quartile of Gene length or GC content and the also, the ratio of h3/polII of DMS by Cond1

correlation and scatterplot with violin and box plot facet by GC content, no bias dettected facet by GC content, no bias dettected correlation of ratio H3_DMSO/PolII_DMSO by ratio H3_Cond1/PolII_Cond1 facet by genelegth, correlation of ratio H3_DMSO/PolII_DMSO by ratio H3_Cond1/PolII_Cond1 facet by GC content, correlation of ratio H3_DMSO/PolII_DMSO by ratio H3_Cond1/PolII_Cond1