SimpleCOMPASS: Fit the discrete COMPASS Model
In RGLab/COMPASS: Combinatorial Polyfunctionality Analysis of Single Cells

Description Usage Arguments Value Note Examples

This function fits the COMPASS model from a user-provided set of stimulated / unstimulated matrices. See the NOTE for important details.

SimpleCOMPASS(
  n_s,
  n_u,
  meta,
  individual_id,
  iterations = 10000,
  replications = 8,
  verbose = TRUE,
  seed = 100
)

`n_s`	The cell counts for stimulated cells.
`n_u`	The cell counts for unstimulated cells.
`meta`	A `data.frame` of metadata, describing the individuals in the experiment. Each row in `meta` should correspond to a row in `data`. There should be one row for each subject; i.e., one row for each element of `n_s` and `n_u`.
`individual_id`	The name of the vector in `meta` that denotes the individuals from which samples were drawn.
`iterations`	The number of iterations (per 'replication') to perform.
`replications`	The number of 'replications' to perform. In order to conserve memory, we only keep the model estimates from the last replication.
`verbose`	Boolean; if `TRUE` we output progress information.
`seed`	A seed for the random number generator. Defaults to 100.

A list with class COMPASSResult with two components, the fit containing parameter estimates and parameter acceptance rates, and data containing the generated data used as input for the model.

n_s and n_u counts matrices should contain ALL 2^M possible combinations of markers, even if they are 0 for some combinations. The code expects the marker combinations to be named in the following way: "M1&M2&!M3" means the combination represents cells expressing marker "M1" and "M2" and not "M3". For 3 markers, there should be 8 such combinations, such that n_s and n_u have 8 columns.

 set.seed(123)
n <- 10 ## number of subjects
k <- 3 ## number of markers

## generate some sample data
iid_vec <- paste0("iid_", 1:n) # Subject id
data <- replicate(2*n, {
nrow <- round(runif(1) * 1E4 + 1000)
ncol <- k
vals <- rexp( nrow * ncol, runif(1, 1E-5, 1E-3) )
vals[ vals < 2000 ] <- 0
output <- matrix(vals, nrow, ncol)
output <- output[ apply(output, 1, sum) > 0, ]
colnames(output) <- paste0("M", 1:k)
return(output)
})

meta <- cbind(iid=iid_vec, data.frame(trt=rep( c("Control", "Treatment"), each=n/2 )))

## generate counts for n_s, n_u
n_s <- CellCounts( data[1:n], Combinations(k) )
n_u <- CellCounts( data[(n+1):(2*n)], Combinations(k) )
rownames(n_s) = unique(meta$iid)
rownames(n_u) = rownames(n_s)
## A smaller number of iterations is used here for running speed;
## prefer using more iterations for a real fit
scr = SimpleCOMPASS(n_s, n_u, meta, "iid", iterations=1000)