1 Interpretation

Do the enrichment results support conclusions or mechanism discussed in the original paper?

• Yes and no. All the mechanisms from the original paper seem there (maybe except for the down-regulated ones) but they don't say anything about TPO signaling which is my biggest cluster.

How do these results differ from the results you got from Assignment #2 thresholded methods?

• Super different. Way more significant results, way fewer total results. Also much bigger focus on TPO instead of apoptosis and immune dysregulation.

Can you find evidence, i.e. publications, to support some of the results that you see. How does this evidence support your result?

• All of it is supported. DS is really well studied. I can find multiple studies directly backing up everything.

2 Pick a next step

Using your networks and results from the previous section add one of the following:

• Add a post analysis to your main network using specific transcription factors, microRNAs or drugs. Include the reason why you chose the specific miRs, TFs or drugs (i.e publications indicating that they might be related to your model). What does this post analysis show?

• No. post analysis with other genes and stuff doesn't really work well with my dataset because its too big.

• Choose a specific pathway or theme to investigate in more detail. Why did you choose this pathway or theme? Show the pathway or theme as a gene network or as a pathway diagram. Annotate the network or pathway with your original log fold expression values and p-values to show how it is effected in your model. (Hint: if the theme or pathway is not from database that has detailed mechanistic information like Reactome you can use apps like GeneMANIA or String to build the the interaction network.)

• Could do with TPO...

• Sometimes the most interesting information is the gene that has no information. In this type of pathway analysis we can only discover what we have already described previously in the literature or pathway databases. Often pathways found in one disease are applicable to other diseases so this technique can be very helpful. It is important to highlight any genes that are significantly differentially expressed in your model but are not annotated to any pathways. We refer to this set of genes as the dark matter.

  • Include a heatmap of any significant genes that are not annotated to any of the pathways returned in the enrichment analysis.
  • Include a heatmap of any significant genes that are not annotated to any pathways in entire set of pathways used for the analysis.

• Could be interesting, but also time consuming...

• I'll do this one, its mostly coding.

• It looks like my genes are either in a pathway returned from the enrichment analysis, or have no annotation. The 2 heatmaps must contain the exact same genes (both 4255 genes, and one is a subset of the other) EVEN WHEN LOOKING AT ONLY SIGNIFICANT PATHWAYS

• I guess this is kind of expected. I'm looking at DS which affects a ton of genes, and is quite well studied. So I guess all the genes are pretty much either part of a well studied, significant pathway, or we don't really know what they do.

• uh oh... i dont think any of my significantly differentially expressed genes are unannotated.

• I guess i'll just taky my most significant of the unannotated?

• NVM I was doing it wrong - there are 323.

• They cluster perfectly by diagnosis and DS is generally upregulated

• Maybe they're just random chr21 genes? I'll check.

• Nope, not even most of them are chr21, theres a big distribution.

• I messed this up - ended up normalizing by chromosome size and chr21 is wayyyy overrepresented :)