search

kstreet13 / slingshot

Functions for identifying and characterizing continuous developmental trajectories in single-cell data.
259 stars 42 forks source link

Slingshot() with treatment conditions and variance in cell cluster #175

Closed DennisFeige closed 2 years ago

DennisFeige commented 2 years ago

Hi @kstreet13,

I have a sample where I have a control (gray) and treatments with three different concentrations. Low (light blue), middle (blue) and high concentration (dark blue). Similar to your tutorial ("Trajectory inference across conditions: differential expression and differential progression") I want to fit a curve(s) to see the distribution of pseudotimes over conditions. I already subset my cells and would like to avoid further subsetting of my cells. The problem is that I can clearly see in the umap that there is a gradient from control to highest concentration with the other concentrations laying roughly on the gradient but I still have variance in my cell cluster "going orthogonally" to my treatment gradient (I hope you understand what I mean). So just fitting a principle curve doesnt work for me. Do you have a good way of how to deal with this? Eg. fitting multiple curves on the same umap that run in parallel?

slingshot Thanks for every help!

Best, Dennis

kstreet13 commented 2 years ago

Hi @DennisFeige ,

Hm, that's an interesting issue. It almost looks like there are two separate trajectories, each containing two conditions. If you're sure that isn't true biological signal, I would actually go back to the normalization/dimensionality reduction steps to remove it. If you treat the different conditions as batches, there are some good adaptive methods for batch integration that attempt to remove this type of difference while preserving as much biological signal as possible.

Specifically, I've found that I get good results with fastMNN, which is what I've used in my published work. I have also tried Seurat integration and Harmony for this type of correction.

Hope this helps! Kelly

DennisFeige commented 2 years ago

Hi @kstreet13,

thanks for the quick answer! I am sure that this cluster contains several biological subclusters which I can identify by marker genes. The thing is that I would like to run this analysis also on other clusters which also have subclusters. So I would like to avoid subsetting each of those to not make it to complicated and keep the biological message clear.

Best wishes, Dennis