Applying cNMF on semi-supervised scRNA clustering

[liangxs7015]

Hi,

Thank you for creating such a wonderful tool! I am currently working with scRNA-seq data consisting of 8 million cells and I am particularly interested in analyzing 100 specific genes. I would like to use cNMF to perform semi-supervised clustering to understand how these genes might affect cell status. This means that I will extract the 100*800,000 matrix from the orignial 50,000 * 800,000 matrix as an input.

Could you please let me know if this application is feasible using cNMF?

Thank you very much for your assistance!

[dylkot]

Hi, I don't fully understand what you mean by semi-supervised clustering in this context. In general cNMF doesn't do "hard clustering" in the usual sense, it outputs continuous scores for each program for each cell. Is that what you are looking for? There shouldn't be any issue with only inputing an 800K x 100 matrix and setting the number of highly variable genes to use to 100.

[liangxs7015]

Hi, I don't fully understand what you mean by semi-supervised clustering in this context. In general cNMF doesn't do "hard clustering" in the usual sense, it outputs continuous scores for each program for each cell. Is that what you are looking for? There shouldn't be any issue with only inputing an 800K x 100 matrix and setting the number of highly variable genes to use to 100.

Thank you for your quick reply. By 'semi-supervised,' I mean that the input genes for the matrix were taken from a predefined gene list. My goal in running this was to observe how the cells would be divided into different states according to the given gene list.

Thanks to your suggestion, I have successfully obtained five GEPs and plan to classify all cells from different datasets into five states based on their highest GEP score in "usage". I'm still working on it and am not entirely sure if this strategy is reasonable. Sorry for the delayed response.