csDeconv: Improve reference-free deconvolution using cross-cell type...

In ziyili20/TOAST: Tools for the analysis of heterogeneous tissues

Improve reference-free deconvolution using cross-cell type differential analysis

Description

This function improve the feature selection in reference-free deconvolution through cross-cell type differential analysis

Usage

csDeconv(Y_raw, K, FUN, nMarker = 1000,
InitMarker = NULL, TotalIter = 30, bound_negative = FALSE)

Arguments

Y_raw	A G*N matrix, G is the number of features, N is the number of subjects; or a SummarizedExperiment object.
K	The number of cell types. Need to be specified a priori.
FUN	The reference-free deconvolution function, this function should take Y_raw and K, and the return values should be a N by K proportion matrix. N is the number of samples and K is the number of cell types. Default function is a wrapper of the RefFreeCellMix() function from CRAN package RefFreeEWAS.
nMarker	The number of markers used in the deconvolution. Default is 1000.
InitMarker	A vector of length L to represent the selection of inital markers. L should be equal or smaller than G. If G is large, it is recommended that L is much smaller than G. If not specified, the most variable nMarker features will be used.
TotalIter	The total number of iterations of applying cross-cell type differential analysis. Default is 30.
bound_negative	Whether to bound all negative parameter estimators to zero.

Value

allProp	A list of estimated proportions from all iterations.
allRMSE	A vector of root mean squared errors (RMSE) from all iteratoins.
estProp	A N*K matrix representing the mixture proportions of K cell types in N subjects, chosen from allProp with the smallest RMSE.
updatedInx	Selected variable index from the algorithm.

Author(s)

Ziyi Li <[e-mail]>

References

Ziyi Li and Hao Wu. "Improving reference-free cell composition estimation by cross-cell type differential analysis".

Examples

Y_raw <- abs(matrix(runif(10000*20, 0,1),10000,20))
K <- 3

## wrap your reference-free
## deconvolution method into a function
## this function should take Y and K as input
## and output a N by K proprotion matrix
## here we use RefFreeCellMix() as an example
outT <- csDeconv(Y_raw, K)

RefFreeCellMix_wrapper <- function(Y, K){
     outY = myRefFreeCellMix(Y,
     mu0=myRefFreeCellMixInitialize(Y,
     K = K))
     Prop0 = outY$Omega
     return(Prop0)
}

outT <- csDeconv(Y_raw, K,
            FUN = RefFreeCellMix_wrapper)

ziyili20/TOAST documentation built on Aug. 28, 2022, 11:28 a.m.