mixOmicsCaret::get_mixOmics_spls() with binary outcome

Analysis
1

Hi all,

I am analyzing a microbiome dataset for feature selection with a binary outcome and came across this tutorial using mixOmics with caret (mixOmics with caret examples). This example closely aligns with my current work.

However, I was confused about why, when training a model using caret, the binary (categorical) outcome is converted to a numeric variable. Additionally, after tuning the parameters, the performance plots display RMSE and Rsquared instead of Accuracy, which is typically used for binary outcomes.

Is there a specific reason for transforming a binary outcome into a numeric one? Does it make interpretation easier, or is it a requirement for analysis? Alternatively, should I be using sPLS-DA instead of mixOmicsCaret::get_mixOmics_spls() within caret::train()? I looked for an implementation of sPLS-DA compatible with Caret package but couldn’t find one.

I would appreciate any insights on this.

Thank you,
Jessica

2

Hi @jessicah,

If you are have microbiome data with a categorical outcome and are interested in feature selection to identify which variables best distinguish your outcome I would definitely recommend the mixOmics model sPLS-DA. You can use this directly using the mixOmics package (see case study tutorial here). Unfortunately I am not familiar with the mixOmicsCaret package so I can’t comment on that but perhaps other mixOmics users are or you could reach out to that package’s maintainer directly.

Hope that helps!
Cheers,
Eva

1 Like