RNA-seq: Can we obtain distances between samples of different experiments?

Question

[salamandra]

Imagine RNA-seq data from three independent experiments, each experiment corresponds to one type of cancer cell line treated with a drug at different time points.

Is it correct to join all this data (as normalized read counts) in a heatmap and then do a dendrogram, to see if the different cell types at the same time point cluster together?

I know this is possible if data from different cell lines were taken from the same experiment, but not I'm not sure if we can compare distances between samples of different experiments.

Also, a naive question: Is a gene more up-regulated if it has a higher fold-change or if it has a lower p-value/FDR and positive fold-change?

Answer 1

[h.mon]

It is possible to compare data from different experiments (e.g. Comparison of the transcriptional landscapes between human and mouse tissues), but then again, it is possible your results would be very difficult to interpret and could lead to misleading conclusions (e.g. A reanalysis of mouse ENCODE comparative gene expression data).

A gene is up-regulated if it has positive fold-change, and down-regulated if it has negative fold-change when comparing a pair of conditions (most usually, control vs treatment). So keep in mind if a gene is up-regulated in e.g. control, it means this gene is down-regulated in treatment. One usually considers a gene to be significantly up- or down-regulated if it fails the null hypothesis of difference in expression between treatments being equal to zero. By convention, thresholds of FDR < 0.05 or FDR < 0.01 or even FDR < 0.001 are used.