search

rpmccordlab / SMILE

SMILE: Mutual Information Learning for Integration of Single Cell Omics Data
11 stars 6 forks source link

single point for each cell in the UMAP #3

Closed imerelli closed 2 years ago

imerelli commented 2 years ago

Hi, I'm using SMILE for integrating different single-cell omics and I find it very useful. One thing that biologists are asking me is to have in the UMAP a single point representing for each cell the all the different omics together. Do you think this make sense? In case, do you have any suggestion about have to proceed?

rpmccordlab commented 2 years ago

The idea we have to let a single point represent all different omics together is, for example, taking the mean of RNA embedding and ATAC embedding.

If your data is joint-profiled, each cell will have a RNA embedding and ATAC embedding. So, you can average its embedding learned by SMILE, and visualize the whole joint-profiled data in UMAP, in which each dot compresses information from RNA and ATAC.

If your data is not joint-profiled, then you may infer ATAC information for a RNA cell through kNN or infer RNA information for a ATAC cell. After SMILE integration, you should be able to find k ATAC neighbors for a RNA cell, and vice verse. Then, you have each RNA cell with inferred ATAC information and visualize your data with UMAP.

We hope this idea would help.