How to choose the proportional variance (p)

[sanderslhc]

Hi, thank you to provide such a convenient pipeline for deconvolution. However, I have some questions about parameter selection. The most important thing is, I noticed we need to set the proportional variance to control the noise, but there is no description about how to choose the proportional variance value. Could you give me some suggestions? Thank you.

[sanderslhc]

Another question is I didn't find how to split the cell type with cell number between 50-100, is it still split in half? But it will produce some rare cell types (<50) in both the reference and pseudo bulk. How to solve this problem?

[annavpo]

Hello! Apologies for delayed reply. First, proportional variance is a parameter that takes effect only for mode=2 for pseudobulk generation. When mode =2 and sampleCT=TRUE : then propVar has to be a positive number and the proportions are drawn from a binomial distribution with mean1 = 0 and sd1= 0.0001 and mean2=1 and sd2=propVar more specifically : rbinorm(Ncelltypes, mean1=0, mean2=1, sd1=0.0001, sd2=propVar, prop=0.5) where n = the number of cell types and proVar controls the standard deviation of the selected cell types. The larger the propVat the more variable the proportions in the samples (higher noise). When mode=2 and sampleCT=FALSE, the propVar can be either a positive number and this time proportions are drawn from a uniform distribution round(runif(nlevels(T_prep@meta.data$cellType), 100, 100 + (propVar*2)))/100 in this case again the larger the propVar the more noise. A low propVar e.g 10 will give very similar proportions in the sample e.g in a sample of 4 cell types it would be ~25%, 25%, 25%, 25%. Generally we use as a default propVar=500 in the pipeline. If proVar is set to a negative number then we use a uniform distibution with set min=1 and max-99: runif(n, min = 1, max = 99). For more details you can have a look at : https://github.com/Papatheodorou-Group/CATD_snakemake/blob/main/Modules/Psuedobulk/randomGen_seeded.R

[annavpo]

Another question is I didn't find how to split the cell type with cell number between 50-100, is it still split in half? But it will produce some rare cell types (<50) in both the reference and pseudo bulk. How to solve this problem?

For cell types with cells between 50-100 we still split them in half although it can produce rare cell types because it is still interesting to see if such low proportions can be detected from the methods. However we keep cell types that have less than 10 cells which we flag as riskGroup and keep all of their cells. For more look here at line 18 and 22: https://github.com/Papatheodorou-Group/CATD_snakemake/blob/main/Modules/Convert_split/split.R if you would like to keep all the cells of a cell type ~50-100 in both the reference and the pseudo-bulk then you can increase the number in line 18. Hope it helped!