Search
Question: Global Trend Analysis using results from EdgeR
0
6 months ago by
anupriya.dalmia0 wrote:

I have generated three DGE lists using the MakeContrasts function in EdgeR for pairwise comparisons of 3 conditions. I now want to make a union of these genes and cluster the expression (preferably, using the k-means algorithm). Following that, I would like to plot the expression profiles of the clusters for the different conditions.

I attempted doing this using ConsensusClusterPlus but I am unable to interpret the tracking plot.

I have a total of 625 DE genes and 532 samples.

Any suggestions?

modified 6 months ago by Aaron Lun21k • written 6 months ago by anupriya.dalmia0
0
6 months ago by
Aaron Lun21k
Cambridge, United Kingdom
Aaron Lun21k wrote:

Your question does not have anything to do with edgeR. As far as I can see, edgeR has given you some result from the pairwise comparisons, and the real issue is how to visualise the differences between samples.

In this regard, I think your current approach is silly. If you only use the genes defined as DE between pre-defined conditions for clustering or dimensionality reduction, you will favour/force your clusters to be equal to your initial conditions! In that case, you might as well just make up a plot with three blobs of points corresponding to the known conditions, as you're forcing the algorithm to give you the expected result.

If you want an unbiased visualisation, I would suggest using plotMDS, which does not use any information about the conditions. If you end up with three clusters of samples corresponding to the conditions, this provides a semi-independent confirmation that your conditions are well-separated. Alternatively, you could also compute log-CPMs with cpm, transpose the result and use it in prcomp (for PCA) or Rtsne (for t-SNE).

ADD COMMENTlink modified 6 months ago • written 6 months ago by Aaron Lun21k