Performing Multimodal and Batch Integration simultaneously

[jules-samaran]

Hi,

Thank you very much for sharing your work, I found it very interesting. When looking at the examples on your repo something confuses me, is it possible to integrate scRNA-seq and scATAC-seq with MultiMAP while accounting for batch effects in each modality at the same time? What I mean is that through one integration procedure you would get embeddings of the cells that would preserve the celltype information while being completely mixed with respect to the batch and modality of origin of each cell. I thought this is what you had done in the Figures 3-e of your genomebiology paper but I don't see how I should use the Multimap.Integration() and Multimap.Batch() methods to do this simultaneously.

Additionally, to evaluate such an integration procedure, would you recommend using the MultiGraph instead of the 2 dimensional output?

Cheers,

Jules

[ktpolanski]

Under the hood, MultiMAP.Batch() works like MultiMAP.Integration(). It's just less involved in terms of preparation, taking out the element of manually setting up independent dimensionality reductions. You trade control for ease of plugging it into a standard single-cell workflow.

My gut response is to split this up into all the individual batches you want merged, so multiple ones per modality in the outlined scenario, and then integrate the lot.

@mikasarkinjain should be more familiar with the details of the paper analysis.