fishash is an R package for calling guides in perturbseq data from
UMI counts, based on treating the count matrix as a contingency table.
For each cell barcode and guide barcode, fishash tests how likely
the 2 barcodes are to co-occur in reads. More specifically, it tests
whether the 2 barcodes have an odds ratio greater than 1 using a
one-sided Fisher's exact test. The method also includes a procedure to
correct for hidden confounding due to Simpson's paradox, and performs
a block-dependence-aware multiple testing correction (assuming that
tests from different cells are independent, but tests within a cell
are dependent).
See our preprint for a full description of the method.
To install the package, do:
devtools::install_github("jackkamm/fishash")
Fishash will also become available in the next version of Bioconductor (3.24). After Bioconductor 3.24 is released (or if you are on the devel version), you can install it with:
BiocManager::install("fishash")
Given a count matrix counts_mat with guides for rows and cells for
columns, you can assign the guides calling the fishash() function:
library(fishash)
# returns a SummarizedExperiment
res_fishash <- fishash(counts_mat)
# for the first few cells, print whether they received 0, 1, or 2+ guides:
head(colData(res_fishash)$demux_type)
# print the assigned guides for the first few cells
head(colData(res_fishash)$assignment)
For more options, see the help page:
help(fishash)
See the vignette for an example on how to use the package.