Can we use total gene single cell data to generate the simulated spatial data

[xiangpin]

Thanks for the work, In the tutorial for cell-type deconvolution, the top marker gene was selected firstly. I have tried to use the total genes to generate the simulated data. But the result was weird, and it was very slow. So I want to whether we can use the total genes of single cell to simulate?

[SONGDONGYUAN1994]

Hi Xiangping, you can use the total genes, BUT I do not recommend it. Here is the problem: the deconvolution result is heavily affected by the choice of cell-type marker genes (in my experiment, the effect of choosing marker genes is much larger than that of choosing different deconvolution methods). Therefore, for the simulation part, I would suggest for the estimation of spot-level cell-type proportion, you may only use marker genes to do so. For the next step, generating the simulated data, you can still generate all genes since it will not related to markers anymore; now you have the spot-level cell type proportion and you just need to simulate whatever cells and mix them in each spot.

Please email me if my explanation is not clear and we can schedule a further discussion.

Best,
Dongyuan