Skip to content

Latest commit

 

History

History
229 lines (177 loc) · 7.8 KB

config_examples1.md

File metadata and controls

229 lines (177 loc) · 7.8 KB
Raw

Back to run DE.

Config examples

Set contrasts

In the config file under contrasts: contrast_list: you can define a list of contrasts that should be computed.

E.g.

contrast_list:
  - title: "A vs C"
    ratio:
      column: condition
      numerator: A
      denominator: C
  - title: "B vs C"
    ratio:
      column: condition
      numerator: B
      denominator: C
    ranking_order: "-x"
    results_parameters:
      altHypothesis: less

defines two contrasts "A vs C" and "B vs C". "A", "B" and "C" are levels of a column "condition" in the covariate file. In the second contrast, also defaults for ranking_order and results_parameters are overwritten. The defaults are defined in the config file under contrasts: defaults:.

conditional functional annotation analysis

Analysis of e.g. GO and KEGG annotation enrichment can be optional for certain contrasts. For example by adding a key-value pair goseq: true to a contrast specification, goseq is run for that contrast for GO and KEGG enrichment analysis.

At the level of the snakefile, the argument if_set = dict(goseq=True) of the expand_path() path handler function is used when collecting output files for the all rule. While expanding over the contrasts, only output files will be requested, if a dict(goseq=True) is in the definition of the respective contrast. This mechanism can also be used, when adding more analysis steps to the pipeline that are optional for contrasts.

If results of some analysis steps are not computed for all contrasts, also the report must adapt by not including sections for those steps for a given contrast. See adding new snippets on how to make snippet inclusion conditional.

included functional annotation methods so far:

  • goseq for GO and KEGG over-representation analysis
  • cluster_profiler for MSigDb, GO and/or KEGG over-representation or gene set enrichment analysis

To run both goseq and cluster_profiler on a contrast, e.g. set:

contrast_list:
  - title: "A vs C"
    vector: [1,1,0,1]
    goseq: true
    cluster_profiler:
      run: true
      MSigDb:
        categories: [H, C2]
        type: gsea
      GO:
        ontologies: [MF]
        type: ora
        pval: 0.1
      KEGG:
        type: ora
        kegg_organism_code: mmu
      KEGG_modules:
        kegg_organism_code: mmu

For cluster_profiler more parameters can be configured. If any of MSigDb, GO, KEGG (pathways), KEGG_modules (modules) is set, the analysis will be run for the respective annotation. All parameters have defaults except for kegg_organism_code, which has to be provided. To use all defaults, e.g. for GO, set GO: {}. Separate tests will be performed for the categories/ontologies provided in the list. If categories/ontologies is not set, all categories/ontologies will be tested.

Configure report

In the config file under report: you can define a 'building plan' how to assemble the report from Rmd snippets.

E.g.

report:
  report_snippets:
    - Covariate_table.Rmd
    - NormalisationQC:
        - PCA_plot.Rmd
        - SampleSimilarity_plot.Rmd
    - contrast:
        - __list__: __contrasts__
  defaults:
    contrast_list:
      - Init_code.Rmd
      - MA_plot.Rmd
      - Result_table.Rmd

Will first insert the snippets (from the report/ directory) in this order:

  • Covariate_table.Rmd

and from sub-folder NormalisationQC/:

  • PCA_plot.Rmd
  • SampleSimilarity_plot.Rmd

Then, the pipeline will create a section "contrast" and add sub-sections for all (special keyword __list__) contrasts (special keyword __contrast__) defined in the config file. For each contrast, report snippets will be added in this order defined under defaults: contrast_list::

  • Init_code.Rmd
  • MA_plot.Rmd
  • Result_table.Rmd

Instead of using __contrast__, contrasts can also be specified as a list, and defaults can be overwritten:

E.g.:

report:
  report_snippets:
    - Covariate_table.Rmd
    - NormalisationQC:
        - PCA_plot.Rmd
        - SampleSimilarity_plot.Rmd
    - contrast:
      - __list__:
        - "A vs C"
        - "B vs C":
          - Result_table.Rmd
  defaults:
    contrast_list:
      - Init_code.Rmd
      - MA_plot.Rmd
      - Result_table.Rmd

Here, for the contrast "B vs C" the default snippet list is overwritten and only the snippet "Result_table.Rmd" is used.

Defaults for list entries (created by __list__) have the suffix _list. Defaults for blocks can also be defined and have no suffix. This definition would be equivalent to the previous example:

report:
  report_snippets:
    - Covariate_table.Rmd
    - NormalisationQC: __defaults__
    - contrast: __defaults__
  defaults:
    contrast_list:
      - Init_code.Rmd
      - MA_plot.Rmd
      - Result_table.Rmd
    contrast:
      - __list__:
        - "A vs C"
        - "B vs C":
          - Result_table.Rmd
    NormalisationQC:
      - PCA_plot.Rmd
      - SampleSimilarity_plot.Rmd

Here, contrast: __defaults__ would load contrast: from the defaults. There, in turn, via __list__ the defaults from contrast_list: would be loaded for the entry "A vs C".

Note: blocks do not have to contain a __list__, they may simply be groups of snippets that are inserted under a common subsection.

Tip: when configuring the report, it can be helpful to copy, paste and edit the report configuration from the default yaml file.

Merging reports of several analyses

You might want to perform different analyses of the same data, using e.g. different design formulas or filtering. To this end, it is possible to create a merged report of several analyses.

Assume for example that results were stored using these two out_path_patterns:

previous_analysis/{contrast}/{step}/out/{step}.{contrast}.{extension}
current_analysis/{contrast}/{step}/out/{step}.{contrast}.{extension}

If the second path pattern is the current path pattern, you may add the results of the first analysis to its report with the line:

report:
  merge: previous_analysis/{contrast}/{step}/out/{step}.{contrast}.{extension}
  report_snippets: ...
  defaults: ...

Then all snippets will be inserted twice, alternating between results of the two analyses. Tags will be added to the sections: "-- analysis0" for the current analysis and "-- analysis1" for the added analysis. Suffixes "_analysis<x>" will also be added to relevant variables in the report, to load and use the corresponding config files and lists of generated results files. Therefore, if you wish to replace the tag in the section names, do a replace with e.g. "-- analysis0" by "-- <my title>" including the "--".

The value of merge can also be a list or a dict. Especially if you don't need all snippets inserted twice, you can select which ones to add:

report:
  merge:
    pre_analysis: previous_analysis/{contrast}/{step}/out/{step}.{contrast}.{extension}
  report_snippets:
    - Covariate_table.Rmd
    - NormalisationQC:
        - PCA_plot.Rmd
        - SampleSimilarity_plot.Rmd
    - contrast: __defaults__
    - pre_analysis:
        contrast: __defaults__

Here, only the section contrast is added twice to the report, once with the tag analysis0 for the current analysis and once with the tag pre_analysis for the previous analysis. In general, the value of merge can be a dictionary with multiple keys and either strings or lists as values.

Note: using tags like pre_analysis: inside of blocks like contrast: is not recommended, since some global variables might be defined there, that are used across different snippets. The values of such variables might differ between analyses, though.

Note: combining path patterns with differing wildcards in one report was not tested yet and might result in error messages.

Back to run DE.