Hi Ziyi Li
I tried to run deconvolution on a set of samples using Tsisal. I was surprised to notice that even when I used the same assumed number of cell types, I got very different results when running deconvolution several times (correlation between estimated cell proportions as low as 0.2 - 0.4). Is it to be expected or am I doing something wrong? If so, what would be the approach, risking minimal bias?
Best regards
Edit:
I think I found out what happened: the differences between runs occur, but they are indeed very small. The initial poor correlation came from the fact that the assumed cell types are assigned different names (numbers). After running a correlation analysis for all cell types between two deconvolution runs I was able to identify cell types with very high correlation.
I edited the original question to remove unnecessary information. Works great for me now, thanks :)
Best regards
Irena
Hi Ziyi Li I tried to run deconvolution on a set of samples using Tsisal. I was surprised to notice that even when I used the same assumed number of cell types, I got very different results when running deconvolution several times (correlation between estimated cell proportions as low as 0.2 - 0.4). Is it to be expected or am I doing something wrong? If so, what would be the approach, risking minimal bias? Best regards
Edit: I think I found out what happened: the differences between runs occur, but they are indeed very small. The initial poor correlation came from the fact that the assumed cell types are assigned different names (numbers). After running a correlation analysis for all cell types between two deconvolution runs I was able to identify cell types with very high correlation. I edited the original question to remove unnecessary information. Works great for me now, thanks :) Best regards Irena