search

KevinMenden / scaden

Deep Learning based cell composition analysis with Scaden.
https://scaden.readthedocs.io
MIT License
71 stars 25 forks source link

Merge mouse whole cell + infiltrating immune cell in cancer #114

Closed mHagiw closed 2 years ago

mHagiw commented 2 years ago

Hi Kevin.

I want to know the percentage of a subset of immune cells in tumor cells in mouse.

So, I'm thinking of combining scRNA-seq data of mouse whole body cells and immune cells isolated from tumors in mouse cancer transplantation models to make simulate data and make predictions.

Do you think this method is possible from a producer's point of view?

This simulate data have about 100 cell types.

May I have your suggestion on this issue? Thanks a lot!

Masaki

KevinMenden commented 2 years ago

Hi Masaki,

it certainly sounds feasible. But I cannot tell how you good this will work in the end, as I haven't tested this with Scaden (or any other deconvolution method).

Generally all I can tell you (unfortunately) is that the training dataset/tissue should be as close as possible to the target dataset. With every little bit you go away from that, you will loose some precision. And unfortunately, currently Scaden isn't able to tell. I wanted to add uncertainty measurements to Scaden for a long time now but didn't find the time so far - maybe this winter I can add it. This would then allow you to do these kind of experiments and hopefully Scaden could tell you how much you can trust the results.

So if it's not too much work for now I would say try it out, maybe also user another deconvolution algorithm on top. If both methods agree to some extent, you can probably trust your results.

Good luck! Kevin

mHagiw commented 2 years ago

Thank you for your reply.

As you said, we will try to match the cell subset of dataset with the bulk data.

Also, another question for the same test, would it be more accurate to define the cell types without being too specific?

For example, I have been dividing CD8 T cells into Tcm, Trm, Tex and so on, but I have the impression that I cannot divide them accurately.

Masaki

KevinMenden commented 2 years ago

Hi Masaki,

sorry for the late reply, I was busy and then on holiday.

Yes definitely, dividing T cells too much will result in poor performance. At some point the difference between the cell types gets lost in all the noise.

Best, Kevin

mHagiw commented 2 years ago

Hi Kevin.

Thank you very much for kind reply.

I will try according to you say !!

Best, Masaki