Reproducible Research

Martin Morgan
February 3, 2015

Importance of reproducibility

A cautionary tale

Facilities enabling reproducible analysis in R

Basics – quickly becomes unsatisfactory

• 'Old-school' text scripts
• Comments describing approaches
• Package versions

Vignettes – 'literate programming'

• Markdown
• R markdown; weaving and tangling
• LaTeX

More advanced considerations

Version control

• Alternative (better!) to using cryptic file names to indicate different versions
• Create a 'repository', add / edit files, 'commit' changes
• Easily review 'diff'erences, restore previous versions, …

• RStudio

  • File -> New Project –> New Directory –> Empty Project. Click 'Create git repository.
  • File –> New File –> R Script. Edit.
  • 'Git' icon on menu bar –> Commit

Packages

• The Rstudio 'wizard'

• Benefits

  • Code re-use standardizes analysis
  • No need to copy / paste code
  • Easy to share with colleagues (work group, company, world)

Lab

Produce a short 'vignette' summarizing your RNA-seq work yesterday. The code chunks might be along the lines of

library(airway)
library(DESeq2)

data(airway)

dds <- DESeqDataSet(airway, design = ~ cell + dex)
dds$dex <- relevel(dds$dex, "untrt")
dds <- DESeq(dds)
res <- results(dds)

Embed this in textual description with relevant descriptive information (title, author, date) as well as text describing nuances of each step, plus figures or tables of your choosing to illustrate relevant aspects of the experimental design or results.

Render this as an HTML document to share with your colleagues at home.