search

cole-trapnell-lab / monocle-release

276 stars 116 forks source link

monocle for single cell sequencing #98

Open odagliyan opened 6 years ago

odagliyan commented 6 years ago

Hi I am using monocle for differential gene expression analysis of single cells. I am trying to compare the expression of genes of a cell subpopulation (a cluster in tSNE plot for example) for two different experimental conditions. However, the most different genes between two conditions tend to be the marker genes that define the cell population. I suspect that this is due to difference in number of cells/genes to be considered for each condition. Do I need to equalize the number of cells for both conditions or monocle takes care of that?

dpcook commented 6 years ago

Hey odagliyan,

I'm not 100% sure I understand the situation. Do you mind elaborating a bit? Maybe some sample code too. You don't need to downsample or anything to match cell numbers between conditions. As long as you have a reasonable amount of measurements (cells) per condition (ie. not like 1 or 2 cells), the regression should work just fine.

If I'm interpreting it right, you have a situation like...let's arbitrarily say you have scRNA-Seq of immune cells +/- cytokine stimulation or something, and you want to do differential expression of a subpopulation (eg. NK cells) between those that received or didn't receive cytokine? To do this, I would have a single CDS file containing both conditions--make sure it's normalized all appropriately. Then I would filter to only have the NK cell cluster for both conditions. Then run differentialGeneTest using "~Condition" as the full model.

muktad commented 6 years ago

How do you load different experimental conditions as a single CDS from 10X using load_cellranger_matrix function when each experimental condition is saved in a single output directory?

Xiaojieqiu commented 6 years ago

thank you @dpcook to @muktad , sorry for the late reply. you can try to load each condition separately and then merge them in R to create a single CDS. Then you should be able to do what David suggested. Also, I would like to point out that an improve differential gene expression will be released soon and it should able to learn more interesting genes than merely condition specific genes.