Genome-wide vs. single-cell

[TJonCooper]

From the vignette + documentation, it's not entirely clear to me what the difference between distributed = "genome-wide" and distributed = "single-cell" is. What is changing? I note that for the SeuratWrapper vignette (https://htmlpreview.github.io/?https://github.com/satijalab/seurat-wrappers/blob/master/docs/cogaps.html) "genome-wide" is used despite handling scRNA-seq data - is this correct?

[jeanettejohnson]

genome-wide learns patterns over genes/rows, single-cell learns patterns over samples/columns. @genesofeve is this correct?

[rossinerbe]

It just changes how the data is split for parallel processing. If genome-wide the data is split into pieces that contain a group of genes and all of the cells/samples. If single-cell each split contains some of the cells and all of the genes. Generally, if you have more genes than cells/samples you should use genome-wide and vice versa.

[LiuCanidk]

Hi, @jeanettejohnson @rossinerbe ,
does that mean if I applt the parameter as "single-cell", I can use this NMF results to help me classfiy clusters of cell subtypes, just like corresponding some specific cells to all genes, and some cells have specificly higher scores for some genes. It is just like the cell type annotation.