Under Construction

  • Define linear models
  • Select method (voom, edgeR/QLF, limma-trend, vanilla limma)
  • Test against a thresholed with TREAT?
  • Use sample weighting?
  • Choose the assay you want to run analysis on.

Tumor vs Normal in TCGA BLCA Indication

blca.dge <- efds %>% 
  filter_samples(indication == "BLCA") %>% 
  flm_def(assay_name = "rnaseq", # can run on CNV, proteomics, etc.
                 covariate = "sample_type", 
                 numer = "tumor", denom = "normal",
                 batch = "sex",) %>% 
  fdge(method = "voom", with_sample_weights = TRUE)
  caption = "Tumor vs Normal DGE in bladder indication with sample weights")
Differential Expression Results
Top 200 [FDR 0.10, abs(logFC) >= 1.00]

Design: ~ 0 + sample_type + sex

Tested: sample_type: tumor - normal

Tumor vs Normal DGE in bladder indication with sample weights

Tumor vs Normal in TCGA CRC Indication

crc.dge <- efds %>% 
  filter_samples(indication == "CRC") %>% 
  flm_def(covariate = "sample_type", 
                 numer = "tumor", denom = "normal",
                 batch = "sex") %>% 
  fdge(method = "voom", with_sample_weights = TRUE)

How do they compare?

Are the same genes differentially expressed across these two comparisons? That is to say, which genes differentially expressed in bladder tumors are also differentially expressed in colorectal tumors?

Which are different?

Running the fdge-specific compare() function will run the interaction model to identify which genes show (statistically) significant differential expression patterns in the tumor vs normal comparisons across indications.

comp <- compare(blca.dge, crc.dge)

One way to visuaize the comparison between two differential expression results is to plot the log fold changes calculated in one comparison vs the other. Points that fall on the 45° show common differential expression patterns among the two indications. The further away the points come off of the 45°, the more dis-similar they are (these are the pvalues that the compare() function calculates).

The plot below shows the genes that have an FDR <= max_padj in either of the original of the two fdge tests, or in the interaction test run in the compare() function.

Note: in the code below, we are using a very conservative max FDR value (0.0005) in order to minimize the number of points drawn into the document. In a “live” analysis, you would likely want to interact with the comp result more deeply by calling shine(comp) to see the effect of different thresholds, dig through the table of statistical results, etc.

report(comp, max_padj = 0.0005)
Differential Expression Results
Top 64 [FDR 0.00, abs(logFC) >= 1.00]

Design: ~ 0 + .grp. + sex

Tested: .grp.: ( xgrp.tumor - xgrp.normal ) - ( ygrp.tumor - ygrp.normal )

Note also: the compare() functionality, in general, is a very nascent and is a huge work in progress.