Biclustering

[antagomir]

Biclustering is a relatively common modeling task in omics data exploration.

Add biclustering as a visualization technique to miaverse. Note that it has at least two uses:

1) taxa x samples matrix biclustering to identify co-occurring elements on the abundance matrix; might benefit from data transformations but this depends on the specific biclustering algorithm

• Model-based biclustering for overdispersed count data with application in microbial ecology by Aubert et al. (2021)

2) correlation matrix biclustering, for instance taxa x metabolites correlations biclustered to reveal co-occurring elements

• Hintikka et al. (2021)
• Lahti et al. (2013)

At the minimum, add some concrete examples in OMA. Can be its own subsection, close to heatmaps.

If it seems feasible, add support to one of the biclustering packages to miaViz. In practice, a wrapper that allows easy calculation of the biclusters, and visualization on heatmap. We may need to consider later if some of the functionality would better go to other packages (mia?) but this is easier once this has been implemented.

This was has been tested, and is a good start to test and to get idea of biclustering:

• biclust, with toy example

This is a recent one, and should be specifically tailored to microbiome data. Worth testing as well, and might be nice as a default method if it works well:

• cobiclust R pkg

It might be worth looking, if these packages have any added value:

• QUBIC
• rqubic
• BicARE

If there are clear differences, let us consider carefully which one/s to choose for wrappers and examples.

In principle, there could additionally possibility to obtain simulated biclusters as in e.g. FABIA as was done in e.g. Röttjers & Faust (2020). These can be used to validate methods. Should be moved to its own issue, if it seems worth implementing. Not urgent.

In the wrapper function manpages or in OMA, cite at least these publications related to microbiome-related biclustering tasks:

• Hintikka et al. (2021)
• Lahti et al. (2013)
• Model-based biclustering for overdispersed count data with application in microbial ecology by Aubert et al. (2021)

[antagomir]

The example workflow could also contain

1. How to filter rows and cols with very low correlations
2. examples on how to visualize some of the most strongly correlated pairs; for checking purposes

[FelixErnst]

This proposal seams to be heavily method driven and therefore should be split between mia and miaViz, if the method makes sense by itself.

To me it seams to be the same case as for #40

[antagomir]

Support for SE/MAE objects could be provided here for cobiclust package for count assays; and for biclust package for the outputs of getCrossExperimentAssociation i.e. dis/similarity matrices between two distinct assays.

Also possible to start with the examples in OMA, and implement the wrapper later. We can discuss how to prioritize issues.

[antagomir]

biclustering has relatively little general use; closing this for now.