BRIDGES followup workshop (feb 22)

[Aariq]

Followup virtual workshop on multivariate analysis. Focused on big picture stuff like diff between supervised and unsupervised analyses, how to interpret results/plots. Maybe somethign about multiple multivariate datasets (e.g. relationship between soil microbial community and plant community)

• [x] Figure out who the audience is (https://bridges.arizona.edu/)

• [x] Make slides

• [x] Publish slides

• [x] Gather datasets (cupcakes & muffins, muir et al?)

• [x] Make datasets available

• [x] Send reminder (monday)

• [x] Teach it (wednesday)

• [x] Edit webiste

• [x] Edit & post recording

• [x] Send followup email

• [x] Add people to myemma audience

[Aariq]

For 2 hr workshop with not very much prep time I should:

• start with my talk(s) about unsupervised vs. supervised
• add intro about what multivariate / multivariable data is and what the challenges are
• swap PLS-DA with RDA
• add a community ecology example?
• come up with an exercise to basically apply code to new dataset (muir et al?)
• add a lot of links on where to learn more

[Aariq]

Trying to decide what package to use for stats:

MASS pros:

• commonly used in intro stats classes (I think), so probably lots of examples exist
• has CCA and LDA implementations for supervised analyses

MASS cons:

• I think LDA isn't recommended generally anymore?
• Not super familiar with the package

ropls pros:

• Nice-ish looking default plots
• Can do unsupervised and supervised with same UI
• Does hypothesis testing without separate package/function

ropls cons:

• Need to install through bioconductor
• Terrible UI, IMO
• Limited to PCA and (O)PLS(-DA)

vegan pros:

• Fairly all-purpose toolkit for multivariate analyses with a variety of methods
• Plots look ok

vegan cons:

• Super domain specific language makes it confusing to learn unless you're a community ecologist

ade4 pros:

• multipurpose toolkit with many methods
• less domain-specific than vegan

ade4 cons:

• I don't know it at all and don't have time to learn

base R pros:

• prcomp() is super simple and easy to use

base R cons:

• supervised analysis options limited (just manova())
• default plots are hard to read and hard to customize

[Aariq]

Slides: https://cct-datascience.quarto.pub/multivariate-statistics-using-the-right-tool-for-the-job