Meta issue functional analyses

[grst]

Description of feature

Functional analyses are often more easier to interpret than the raw list of differential gene expression analyses.
Let's use this issue to get an overview of use-cases and methods we want to implement.

It's a non-goal to add all possible methods for Gene Set Enrichment Analysis. Instead let's focus on 1-2 methods per use case that implement the current best practice and are computationally efficient.

A general consideration here is if the methods should be executed on the DE results (e.g. GSEA, decoupleR on statistics, ORA ...) or
if they should be executed on the gene expression (e.g. ssGSEA, decoupleR on TPM, ...). The former means the statistics are computed on genes while the latter means statistics are computed on samples.

Gene set enrichment analysis of custom gene sets and or predefined gene sets (e.g. GO, HALLMARK, ...)

• GSVA Add GSVA Gene Score analysis #24
• GSEA add module GSEA #311 + already implemented
• PLAGE Add PLAGE Gene Score Analysis #25
• SingScore Add SingScore Gene Score Analysis #26
• GSEA preranked Allow or switch to GSEA preranked #36
• Over representation analysis (ORA) Biomarker overlaps/ associations with differential gene lists #83
• decoupleR Add module decoupleR to nf-core/modules #350
• gProfiler (already implemented)

Transcription factors

• collecTRI + decoupleR Transcription factor and Pathway scoring using DoRothEA/PROGENy #99 Add module DoRothEA to nf-core/modules #349 Add nf-core module DoRothEA to nf-core/differentialabundance #352

Cancer pathways

• PROGENy + decoupleR Transcription factor and Pathway scoring using DoRothEA/PROGENy #99 Add nf-core module PROGENy to nf-core/differentialabundance #351

cell type deconvolution

• immundeconv for immune cell signatures Create subworkflow for deconvolution analyses in bulk data #361 Create nf-core module for immunedeconv and add to pipeline #368
• deconvolution with custom (often single-cell based) signatures: See https://www.biorxiv.org/content/10.1101/2024.06.10.598226v1 for benchmark of methods

Anyone feel free to suggest other databases and/or methods.

CC @tschwarzl @apeltzer @atrigila @nschcolnicov @alanmmobbs93

[grst]

Notably, decoupleR implements the following methods from the list:

• GSVA
• ORA
• GSEA

[grst]

A general consideration here is if the methods should be executed on the DE results (e.g. GSEA, decoupleR on statistics, ORA ...) or
if they should be executed on the gene expression (e.g. ssGSEA, decoupleR on TPM, ...). The former means the statistics are computed on genes while the latter means statistics are computed on samples.

For us, it would certainly be useful to compute signature scores per sample as we may need to report them to a clinical database. The per-sample scores should be independnet of other samples as we have no control over what subsets of the data may be retrieved by others. SingScore/decoupleR on TPM certainly fulfil these criteria.

On the other hand, statistical power may be reduced when comparing scores between groups. The information from multiple genes is aggregated into a single value, we, therefore, loose the information that changes may be subtle, but coordinated changes into the same direction.

[grst]

Status: wait until @suzannejin and team added a subworkflow for enrichment analyses (#384). Then work on extending it as necessary.