Which transformed plot looks "better" w/ varying cofactors? #95
-
|
Hi there, The theory is very well described in the guideline (https://wiki.centenary.org.au/display/SPECTRE/Data+transformation) regarding which cofactor to use for the arcsinh transformation. Nevertheless, would you mind helping me assess visually which plots (Question 1) looks "better"? Is there sort of check list for what to ensure visually to ensure appropriate transformation, e.g., to have all the cells shown up on the plot, or negative population for a marker does not look too spread out visually, etc.? Thank you again for your help! Question 1.
This marker C seems to need a cofactor even higher than 1,000? Question 2. In the figure below, cofactor = 100 seems to have 2 populations for CD8A, but cofactor = 1000 has these 2 populations merged. Which one in this case is more accurate/better?
|
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 1 reply
-
|
IMHO |
Beta Was this translation helpful? Give feedback.
-
|
Thank you @SamGG very much for your response! Do you in your analysis apply flowScape or OTflow or FlowVS (https://bmcbioinformatics.biomedcentral.com/articles/10.1186/s12859-016-1083-9) in deciding cofactor within the Spectre workflow? |
Beta Was this translation helpful? Give feedback.
IMHO
Q1: 300 sounds OK for all. 1000 is stacking events on the left.
Q2: well known effect of under-estimating the cofactor. A -1000, B 10000 (or a little bit less ~5000, in order to get similar dispersion of the positive and negative populations).
Folcarelli et al. 2021 gives an overview of the field. I like Finak et al. and Azad et al., but directly linked to your question is this figure from Ray et Pyne in 2012.