How to decide on present/non present cell types of each spot?

[maryam2079]

Hi,

Thank you for this great tool.

I am a bit confused on how to interpret the statistical meaning of the cell abundance output of cell2location. My questions are:

1. Is it correct to say that the cell abundance is the estimate of cell counts of that specific cell type for that spot? If yes, can we disregard any cell type that have an abundance below 1? Since counts can not be below 1.

2. In my task, I need to convert the abundance to binary label for each spot. For instance saying that spot S has cell types C1, and C2 in it. How do I decide which cell types are present in each spot based on the abundance?

[J-Anzules]

I worked a lot with the posterior summaries of the model output for cell types (q05, mean, q95). From my understanding, we are given these summary statistics so we can make our own decisions and interpretations.

Since you haven't received any responses, I figured I'd try to help from my experience.

The model tries to predict cell abundances likely to be in that spot in a way that adds up the user-defined cell numbers (This isn't exactly what's going on, but it's close). If you define cell numbers as 20, then I think it's reasonable to say that any mean value for a cell type that is below 1 isn't there. But the model prediction is fluid. For my work, I'm using it on VisiumHD data, and I defined the cell abundance as 1.5. None of my predicted cell types exceeds 0.4 (mean). So, I defined a reasonable cutoff based on the summary statistics for annotating the cells. For you, I would look at the distribution's values in a big-picture way. If you see that certain cell abundance values at some spots are clearly lower than at others, I'd try to set a threshold metric for which cells are present or absent based on the posterior distributions.

Hope this helps!

Happy to talk more about this, I'm trying to wrap my head around this model myself. No one in my cohort has any experience with this model.