run_glmeiv_at_scale_simulatr: Run GLM-EIV at scale simulatr

In timothy-barry/glmeiv: Implements the GLM-EIV model and method

Run GLM-EIV at scale simulatr

Description

Run GLM-EIV at scale simulatr

Usage

run_glmeiv_at_scale_simulatr(
  dat,
  m_fam,
  g_fam,
  covariate_matrix,
  m_offset,
  g_offset,
  alpha = 0.95,
  n_em_rep = 15,
  save_membership_probs_mult = 250,
  pi_guess_range = c(1e-05, 0.03),
  m_perturbation_guess_range = log(c(0.1, 1.5)),
  g_perturbation_guess_range = log(c(0.5, 10)),
  exponentiate_coefs = FALSE,
  ep_tol = 1e-04,
  rm_covariate = ""
)

Arguments

dat	data frame containing columns "m" and "g" for mRNA counts and gRNA counts
m_fam	family describing m
g_fam	family describing g
covariate_matrix	the matrix of covariates; NULL if there are no covariates.
m_offset	offsets for m
g_offset	offsets for g
alpha	confidence level
n_em_rep	number of times to repeat EM algorithm in reduced model
save_membership_probs_mult	save posterior membership probabilities at this multiple
pi_guess_range	range of values over which to sample pi
m_perturbation_guess_range	range of values over which to sample m_perturbation
g_perturbation_guess_range	range of values over which to sample g_perturbation
exponentiate_coefs	(boolean) should the fitted coeficients (and associated standard errors) be exponentiated?
ep_tol	(optional) EM convergence threshold
fam_str	string (currently either "poisson" or "Negative Binomial") giving family object

Value

fitted GLM-EIV model

Examples

m_fam <- g_fam <- augment_family_object(poisson())
n <- 50000
lib_size <- rpois(n = n, lambda = 5000)
m_offset <- g_offset <- log(lib_size)
pi <- 0.01
m_intercept <- log(0.01)
m_perturbation <- log(0.5)
g_intercept <- log(0.005)
g_perturbation <- log(2.5)
covariate_matrix <- data.frame(batch = rbinom(n = n, size = 1, prob = 0.5))
m_covariate_coefs <- log(0.9)
g_covariate_coefs <- log(1.1)
dat <- generate_full_data(m_fam = m_fam, m_intercept = m_intercept,
m_perturbation = m_perturbation, g_fam = g_fam, g_intercept = g_intercept,
g_perturbation = g_perturbation, pi = pi, n = n, B = 2,
covariate_matrix = covariate_matrix, m_covariate_coefs = m_covariate_coefs,
g_covariate_coefs = g_covariate_coefs, m_offset = m_offset, g_offset = g_offset)[[1]]
m <- dat$m; g <- dat$g; p <- dat$p
# ability to recover ground truth given p
fit <- run_glmeiv_at_scale_simulatr(dat, m_fam, g_fam, covariate_matrix, m_offset, g_offset)
fit <- run_glmeiv_random_init_simulatr(dat, m_fam, g_fam, covariate_matrix, m_offset, g_offset)

timothy-barry/glmeiv documentation built on Jan. 30, 2024, 3:46 p.m.