(Note: there is a lot of philosophizing going on here).
It is often the case that an analysis over a set of features (or samples)
induces a ranking over the features (or samples), which is determend by the
test performed in the analysis. The ranks() and signatures() functions
returns a ranking induced over the features (or samples) from the analysis.
ranks(x, ...)A FacileAnalysisResult
When an analysis imposes ranks, this usually only occurs over only one of
"features" or "samples" used in the analysis. In the even that both of these
can be ranked, then these functions will accept a type argument which you
can parameterize with by either "features" or "samples".
Signatures are essentially a summary extracted from the ranks. This is most often the "topn" ranks returned from the analysis (or a dimension thereof).
Let's consider a differential gene expression (DGE) analysis, where we are testing the differential abundance of a gene across two groups of samples. The result of the analysis can induce both a signed and unsigned ranking on the genes under test.
dge <- FacileData::exampleFacileDataSet() %>%
FacileData::filter_samples(indication == "BLCA") %>%
flm_def(covariate = "sample_type",numer = "tumor", denom = "normal",
batch = "sex") %>%
fdge()Ranking the results of the DGE by ascending p-value will provide an unsigned ranking on the genes: Alternatively, one could get a signed ranking from the result of this test simply by arranging each gene by its log-fold-change. Both approaches are achieved by the code below:
rank each gene by its p-value (ascending).
Any differential gene expression analysis works over a set of samples in order to find which genes are most differentially abundant between the two conditions that are defined by the sample groups. It's clear to see, here, that this analysis induces a ranking over the features (genes). This ranking can be both signed or unsigned.
Think about how the next analysis you implement fits this scenario.
Perhaps we can consider an e-/p-/etc- *QTL analysis is on whose features are SNPs, which we can rank by ones that have strong association with the quantitative phenotypes under test.
I feel like being able to generate succinct summaries of an analysis will more easily enable the analyst (via GUI or code) to dive back in and ask another question. That question might be as simple as "how does this version of my question compare to a slightly different version?"