# Getting genes specially up or down regulated [closed]

I have 6 RNA-seq samples like this

4 patients (005, 036, 121, 013)

I have 3 tumour samples and 3 cancer models (organoid)


This is PCA of log transformed data by DESeq2 (tumour Vs. organoid)

In this paper https://www.nature.com/articles/s41467-018-05190-9

They are saying

Finally, for each patient, the genes that are specifically up- or down regulated in each patient’s organoid (1 vs. all comparison, p ≤ 0.05 and abs (log2-FoldChange) ≤ 1) and that are among the top-50 genes with highest base mean expression were selected.

They have tumour and organoid like my case; My confusion is what they mean by "each patient’s organoid (1 vs. all comparison"? Because like them I need up or down regulated genes in each patient's organoid but I can not figure out I should compare each organoid versus what?

The ultimate goal would finding a signature of genes specific to each organoid as each organoid originates from tumour of a patient by these signatures one would be able to classify cancer to sub types

• Nice results! It will be very easy to find the responsible genes with the tightness of those clusters You perform eg. a linear regression (GLM or machine learning) and look at the regression weights (genes). The results are so clean any method of classification will identify them. Just to mention I don't do cancer (never have, never will), but those results look cool.
– M__
Mar 4 '19 at 12:22
• Thanks a lot for your positive and encouraging comments. The assumption is that the expression pattern in each organoid should be very close and similar to the tumour from which this organoid originates. More close better an organoid would be a model of that tumour. But in my PCA we are seeing tumours are in one way and organoids are another way. My expectation was for example tumour 005 and organoid 005 placed close to each other. However I will need a signature of each oganoid as a model of cancer. You meant I take genes as independent variables and a specific organoid as response variable? Mar 4 '19 at 12:31
• @MichaelG. Could you please elaborate on how you would use GLM to do this? Or could you send me the names of some of the packages you have used to do this? I'd like to try it out myself.. Thanks. Mar 4 '19 at 20:11
• @FereshTeh, just came across this rather old post. Regarding "My expectation was for example tumour 005 and organoid 005 placed close to each other" part of your comment above, please see the last paragraph of the "RNA-seq analysis" section of the methods from the paper that you shared. They kick out genes that they think arise from the "artificial environment" (their own words) organoids are prone to. You would get a better looking PCA plot if you would follow their practice, however, organoid biology would still be very very (the nuance is arguable) different from tumor biology.
– haci
Sep 7 '19 at 19:05