A bit about me

Got my Bachelor and Master in Statistics between the Universities of Bologna, Glasgow and Melbourne

A bit of research

• Master research in Population Genetics at the WEHI + worked for 1.5 years as RA
• PhD in Cancer Genomics at the WEHI

A bit outside of research

• After my PhD, I Decided to try a different path outside of research
• Worked as Data Science consultant for 1.5 years at the AI consultancy Eliiza

What I actually do every day

🏡💻 Work in a fun team: Work with a fun interdisciplinary team of scientists, developers, marketing savvy.

📙 Learn: Learn the intricacies and amazingness of mass spectrometry (= most used technique to quantify proteins in a sample) & what life scientists need to make the best use of their experiment.

👩‍💻 Code in R: Assemble workflows in R to analyse mass spectrometry data.

👐 Open Science: Learn and strive for reproducibility and openness in what we produce.

🥜 In a nutshell: Study & Learn, think, build solutions (mainly in R packages)‚ debug, debug, debug, repeat.

Disclaimer

• My highlights also corresponds to talks presented in our timezone

• All talks and workshops will be made available online very very soon! I'll keep you posted

Topics that I will cover

• Teaching and Learning R & Statistics
• Data visualisation
• Data Validation and Package Testing
• Surprise!
• Keynote: Heidi Seibold on Research Software Engineers (RSE) & Academia
• R in Production and Package dependencies
• aRt gallery

Developing a datasets based R package to teach environmental data science

Author and speaker: Allison Horst

• 🔗Introduced the {lterdatasampler}: LTER Data Sampler 📦 (LTER = Long Term Ecological Research program (LTER) Network)
• LTER goal is education and training

Lesson learnt from this talk

• A great way to learn how to build an R 📦 is to create a data-package (package that only includes data)

• I don't have super complex, cool new functions... I cannot write an R 📦

• Free gift! Data packages are an enormously useful tool for teaching purposes (how many times have you used the {iris} dataset [1]??!!)

R. A. Fisher (1936). "The use of multiple measurements in taxonomic problems". Annals of Eugenics. 7 (2): 179–188. doi:10.1111/j.1469-1809.1936.tb02137.x. hdl:2440/15227

🐧 {palmerpenguins} is the new {iris} 🌷

install.packages("palmerpenguins")
library(palmerpenguins)

library(palmerpenguins)
library(dplyr)
library(DT)
penguins %>% head()

## # A tibble: 6 × 8
##   species island bill_length_mm bill_depth_mm flipper_length_… body_mass_g sex  
##   <fct>   <fct>           <dbl>         <dbl>            <int>       <int> <fct>
## 1 Adelie  Torge…           39.1          18.7              181        3750 male 
## 2 Adelie  Torge…           39.5          17.4              186        3800 fema…
## 3 Adelie  Torge…           40.3          18                195        3250 fema…
## 4 Adelie  Torge…           NA            NA                 NA          NA <NA> 
## 5 Adelie  Torge…           36.7          19.3              193        3450 fema…
## 6 Adelie  Torge…           39.3          20.6              190        3650 male 
## # … with 1 more variable: year <int>

Creating R packages: resources that I found super useful!

Get started with 🔗R Packages by Hadley Wickham. Easy to read and very comprehensive.

What you need to get started:

1. Code (one function is enough) AND/OR data (it doesn't have to be large!)

2. An R project in a new folder /path/to/myPackage

3. Run the code usethis::create_package("/path/to/myPackage") (more info https://r-pkgs.org/workflows101.html). This will create the metadata and other files that you need to package the package up!

4. You're setup!

To recapitulate: Setup your R package in a few lines!

usethis::create_package("/path/to/myPackage")

data_fake <- data.frame(First = seq(1:200), 
                        Second = rep("A", 200))
usethis::use_data(data_fake)

{fusen} 📦: Create a package from Rmd

Speaker and author: Sébastien Rochette

If you know how to create a Rmarkdown file, then you know how to build a package.

🔗Introduction to {fusen}

Philosophy You don't need to move around functions and files to create a package, you only need your Rmd with functions, documentation, tests, examples.

install.packages("fusen")
library(fusen)

Rmd first approach to write an R 📦

Building and maintaining OpenIntro using the R ecosystem

Speaker: Cetinkaya-Rundel, Mine

• 🔗OpenIntro website has loads of resources to teach and learn statistics (for free!)
• 🔗OpenIntro R 📦 has supplemental functions and data for OpenIntro resources.

Another lesson to build a data-centric-package!

{OpenIntro} depends on 3 other data packages. See package 🔗DESCRIPTION

Teaching and learning Bayesian statistics with {bayesrules} 📦

Speaker: Mine Dogucu (🔗GitHub repo and 🔗Talk slides )

Keynote: Expanding the vocabulary of R graphics

Speaker: Paul Murrel author of {grDevices} 📦

🔗Paul Murrel's New Features in the R Graphics Engine

Gradient and radial fill

library(grid)

Pattern fills

library(grid)

The {virgo} 📦

Speaker: Stuart Lee

Authors: Stuart Lee and Earo Wang

🔗Introduction to virgo

• Allows to easily build interactive graphics for exploratory data analysis

• Allows cross interactivity between plots without having to build a Shiny app

• virgo plots also works within Shiny

{virgo} in action!

library(virgo)
library(palmerpenguins)

selection <- select_interval()

p <- penguins %>% 
  vega() %>% 
  mark_circle(
    enc(
      x = bill_length_mm, 
      y = bill_depth_mm, 
      color = encode_if(selection, species, "black")
    )
  )

p

🔗Introduction to virgo

p_right <- penguins %>% 
  vega(enc(x = body_mass_g)) %>% 
  mark_histogram(bin = list(maxbins = 20)) %>% 
  mark_histogram(color = "purple", bin = list(maxbins = 20),
    selection = selection) %>% 
  mark_rule(enc(x = vg_mean(body_mass_g)), color = "red", size = 4,
    selection = selection)

p_right

🔗Introduction to virgo

Concatenate interactivity!

hconcat(p, p_right)

🔗Introduction to virgo

Do you see what I see? The {microshades} 📦

remotes::install_github("KarstensLab/microshades")

{microshades} in action!

🐧 {palmerpenguins} with {microshades} example code: https://karstenslab.github.io/microshades/articles/non-microbiome_data.html

{autotest} 📦: Automatic testing for R packages

Speaker and author: Mark Padgham, Software Research Scientst at rOpenSci

🔗Introduction to {autotest}

install.packages("autotest")

• {autotest} goes into the examples of your R 📦 functions and mutates (aka changes) the inputs parameters to function calls.

• This allows to check for robustness of the package to several inputs

{autotest} in actions

library(autotest)
y <- autotest_package(package = "stats", functions = "var", test = TRUE)

A fresh look at unit testing with {tinytest} 📦

install.packages("tinytest")

{tinytest} in action!

🔗Overview of {tinytest} functionalities

library(tinytest)
addOne <- function(x) x + 1
subOne <- function(x) x - 2

# this test should pass
tinytest::expect_equal(addOne(1), 2 )

## ----- PASSED      : <-->
##  call| tinytest::expect_equal(addOne(1), 2)

# this test will fail
tinytest::expect_equal(subOne(2), 1 )

## ----- FAILED[data]: <-->
##  call| tinytest::expect_equal(subOne(2), 1)
##  diff| Expected '1', got '0'

Tutorial: Data validation with the {validate} 📦

Author: Mark van der Loo

The purpose is to provide easy to use tools to check that you're data is valid!

• 🔗Easy data validation for the masses
• 🔗The Data Validation Cookbook

{validate} in action!

library(palmerpenguins)
head(penguins)

## # A tibble: 6 × 8
##   species island bill_length_mm bill_depth_mm flipper_length_… body_mass_g sex  
##   <fct>   <fct>           <dbl>         <dbl>            <int>       <int> <fct>
## 1 Adelie  Torge…           39.1          18.7              181        3750 male 
## 2 Adelie  Torge…           39.5          17.4              186        3800 fema…
## 3 Adelie  Torge…           40.3          18                195        3250 fema…
## 4 Adelie  Torge…           NA            NA                 NA          NA <NA> 
## 5 Adelie  Torge…           36.7          19.3              193        3450 fema…
## 6 Adelie  Torge…           39.3          20.6              190        3650 male 
## # … with 1 more variable: year <int>

table(penguins$island, penguins$species)

##            
##             Adelie Chinstrap Gentoo
##   Biscoe        44         0    124
##   Dream         56        68      0
##   Torgersen     52         0      0

Create a validator with rules

• Separate multiple validations by a comma

• The example shows multivariate validation including completeness validation (is_complete) and conditional validations.

library(validate)
rules <- validator(flipper_length_mm > 0, 
           is_complete(bill_depth_mm, flipper_length_mm, bill_depth_mm), 
           if(island %in% "Biscoe") species %in% c("Adelie"))

confront(penguins, rules) %>% summary()

##   name items passes fails nNA error warning
## 1   V1   344    342     0   2 FALSE   FALSE
## 2   V2   344    342     2   0 FALSE   FALSE
## 3   V3   344    220   124   0 FALSE   FALSE
##                                                     expression
## 1                                        flipper_length_mm > 0
## 2 is_complete(bill_depth_mm, flipper_length_mm, bill_depth_mm)
## 3       !(island %vin% "Biscoe") | (species %vin% c("Adelie"))

Here is the anomalous-down!

What is an RSE?

https://researchsoftware.org/

Heidi’s path

Heidi's Podcast

What does an RSE do?

• An RSE builds software for research

• Generally writes code and teaches about software to researchers

• Consult researcher with any kind of software problem

Tools for Open and Reproducible Science

It's a bit too much to expect that the researcher would do all of those things + their research!

The RSE comes to help!

If you think this does for you...

You can become part of the community!

What does "R in production" mean?

Example from my experience.

• 👩‍💻 I'm a data scientist at Mass Dynamics and I build R 📦 to analyse mass spectrometry data

• But, Mass Dynamics wants to make the functionalities of the R packages easily available also to 👩‍🔬 life scientist, reducing the barrier of having to learn to code

• The solution: the life scientist can interact with an easy user interface (UI, aka frontend) which runs my R 📦 in the background (aka backend)

• Every time a scientist interacts with the UI, the R 📦 is run -> This is R in production 🎉!

How do we do that?

There is a bit of engineering setup and jargon to digest & there are various of way of accomplishing this task!

Tricky aspects:

• A lot of aspects are around the engineering setup (which is not my expertise!)

• However, from my side I need to make sure that:

  • all dependencies needed by my R packages are available in production

  • all packages are put into production with a defined version to allow reproducibility

  • Managing dependencies can be really tricky!

  • How do you know all that you need, when a package will depends on another package, and anothe rpackage will depend on another one etc...

What are package dependencies?

You find them in the DESCRIPTION file of a package, disguising under:

• Depends, Imports, SystemRequirements (dependencies external from R)

• Example from the {sf} 📦 https://github.com/r-spatial/sf/blob/master/DESCRIPTION

How to find them all?

How do you determine all the dependencies needed to reproduce an R package/project environment to make it reporodcible, open, shareable, safe for production?

My summary of suggestions after discussing with speakers at useR! 2021:

• {renv} 📦:

  • is the emerging method to manage package/project dependencies in R.
  • Calling renv::snapshot() saves the state of the project library to the lockfile (called renv.lock)
  • Really useful but for a given project it could give you more than you need (may affect speed).

• Hard code your dependency, start minimal and grow:

  • It's really useful to grow your dependency list little by little.
  • Start with package imports (looking at the DESCRIPTION file) and then look for system dependencies

Ways to find out system dependencies of a 📦

• r-hub/sysreqs 📦 provides a database with API to quickly find out which packages or other software needs to be available to build and use R packages.

  • Usage: sysreqs::sysreq_commands(desc = "path/to/a/DESCRIPTION/file") runs all the commands to install the necessary runtime system dependencies.

• rstudio/r-system-requirements: RStudio independently maintained catalogue of dependencies, used to power the RStudio package manager

• {maketools} 📦 To get runtime dependencies (only for Linux)

maketools::package_sysdeps("stringi")

## # A tibble: 1 × 6
##   shlib          package headers source version url  
##   <chr>          <chr>   <chr>   <chr>  <chr>   <chr>
## 1 libc++.1.dylib <NA>    <NA>    <NA>   <NA>    <NA>

These awesome suggestions from:

Suggestions from speakers at useR! 2021:

• Peter Solymos speaker for Data science serverless-style with R and OpenFaas
• Max Held speaker for: Bridging the Unproductive Valley: Building Data Products Strictly Without Magic

Other resources:

• Interesting blog: Determining system dependencies for R projects
• R-blogger post about useR! 2021 Integrating R into production talks

Productionising ML models developed in R

Author: Surya Avala, ML Engineer at @whispr (Talk slides)

(Not from useR!) But I found it really useful!

Phylosphy: Wrapping up everything with Docker 🐳 and using {plumber} 📦 to generate and API for R.

Thanks to

• useR! organisers and speakers!
• Mass Dynamics for letting me take part to the conference
• Meet Xaringan: Making slides in R Markdown by Alison Hill, learn how to make beautiful slides using {xaringan}.
• These slides were created using the {xaringan} 📦 by Yihui Xie (2021). xaringan: Presentation Ninja. R package version 0.22. https://CRAN.R-project.org/package=xaringan
• Shazia Ruybal Xaringan's slides from her previous R-Ladies Melbourne talk for examples

Glossary of R packages mentioned