Run sceptre in memory

Runs sceptre in memory. This function is appropriate for data of intermediate size (i.e., the data are small enough to fit into memory, and the analysis does not need to be run across nodes on a computer cluster).

run_sceptre_in_memory(
  storage_dir,
  expression_matrix,
  perturbation_matrix,
  covariate_matrix,
  gene_gRNA_pairs,
  side,
  pod_sizes,
  regularization_amount = 0.1,
  seed = 4,
  B = 500
)

Arguments

storage_dir	name of a directory in which to store logs, intermediate computations, and final results
expression_matrix	a matrix of gene expressions (in UMI counts); rows correspond to genes and columns to cells; the rows should be named.
perturbation_matrix	a binary matrix of gRNA perturbations; rows correspond to gRNAs and columns to cells; the rows likewise should be names.
covariate_matrix	a data frame of covariates; rows correspond to cells.
gene_gRNA_pairs	a data frame with named columns "gene_id" and "gRNA_id" giving the names of the gRNAs and genes to analyze.
side	sidedness of the test; one of "left", "right," and "both". "left" is most appropriate for experiments in which cis-regulatory relationships are tested by perturbing putative enhancers with CRISPRi.
pod_sizes	an integer vector giving the size of the "gene", "gRNA", and "pair" pods. sceptre groups the genes, gRNAs, and gene-gRNA pairs into distinct "pods" and runs computations on these pods in parallel. Smaller pod sizes give rise to greater parallelization. pod_sizes should be vector of length three with names "gene", "gRNA", and "pair" and integer values giving the pod sizes.
regularization_amount	(optional; default 0.1) the amount of regularization to apply to the estimated negative binomial size parameters, where 0 corresponds to no regularization at all.
seed	(optional; default 4) seed to pass to the random number generator.
B	(optional; default 500) number of random samples to draw in the conditional randomization test.

Value

a data frame containing the results. Includes columns gene_id, gRNA_id, p_value, skew_t_fit_success (indicating whether the skew-t fit succeeded), xi, omega, alpha, nu (parameters of the skew-t distribution), z_value (the ground truth negative binomial test statistic), and n_successful_resamples (indicates how many of the B resamples were successful).

Examples

if (FALSE) {
# generate random perturbation and expression data in memory
library(dplyr)
set.seed(4)
n_gRNAs <- 50
n_genes <- 40
n_cells <- 5000
# create perturbation, expression, and covariate matrices
perturbation_matrix <- replicate(n = n_gRNAs, rbinom(n_cells, 1, 0.05)) %>% t()
expression_matrix <- replicate(n = n_genes, rpois(n_cells, 1)) %>% t()
covariate_matrix <- data.frame(p_mito = runif(n = n_cells, min = 0, max = 10),
                               lg_umi_count = log(rpois(n = n_cells, lambda =  500)))
# assign column names to the perturbation and expression matrices
row.names(perturbation_matrix) <- paste0("gRNA", seq(1, n_gRNAs))
row.names(expression_matrix) <- paste0("gene", seq(1, n_genes))
# select 5000 random gene-gRNA pairs to analyze; create the gene_gRNA_pairs data frame
gene_gRNA_pairs <- expand.grid(gene_id = row.names(expression_matrix),
                               gRNA_id = row.names(perturbation_matrix)) %>% slice_sample(n = 90)
# let the temporary directory be the storage dir
storage_dir <- tempdir()
# set the remaining parameters: test sidedness, pod_sizes, regularization_amount, seed, and B.
side <- "left"
pod_sizes = c(gene = 10, gRNA = 10, pair = 15)
regularization_amount <- 0.1
seed <- 4
B <- 500
result <- run_sceptre_in_memory(storage_dir,
                     expression_matrix,
                     perturbation_matrix,
                     covariate_matrix,
                     gene_gRNA_pairs,
                     side,
                     pod_sizes,
                     regularization_amount,
                     seed,
                     B)
}