Converting Seurat object to on-disk after normalization

[srangasw-bandolab]

Thank you so much for developing this excellent tool! I was wondering how you would recommend converting a seurat object that already contains normalized data layers (like sacle.data and data) to an on-disk representation with BPCells. Do we need to create a separate matrix directory for each layer? Additionally, the dimensional reduction embeddings can be saved in memory of the data and counts matrices are not being stored in memory, right? Thank you so much!

[immanuelazn]

Hi @srangasw-bandolab, unfortunately we do not have a released version of a BPCells object holder that can hold multiple different assays in BPCells. This is in the works though!

Currently you will have to hold each layer for your Seurat object as a separate BPCells IterableMatrix. Dimension reduction objects are typically much more small, since features typically top out at 50 rather than the thousands in gene/peak/tile space. This means it can fit in your memory without being convereted into a BPCells object most of the time. The counts matrix however, you would save as a disk-backed represenation.