dcats_GLM: A Generalised linear model based likelihood ratio testing

In huangyh09/DCATS: Differential Composition Analysis Transformed by a Similarity matrix

A Generalised linear model based likelihood ratio testing

Description

GLM supports both beta-binomial and negative binomial from aod package.

Usage

dcats_GLM(
  count_mat,
  design_mat,
  similarity_mat = NULL,
  pseudo_count = NULL,
  base_model = "NULL",
  fix_phi = NULL,
  reference = NULL
)

Arguments

count_mat	A matrix of composition sizes (n_sample, n_cluster) for each cluster in each sample
design_mat	A matrix or a data frame of testing candidate factors (n_sample, n_factor) with same sample order as count_mat. All factors should be continous and categorical with only two levels.
similarity_mat	A matrix of floats (n_cluster, n_cluster) for the similarity matrix between cluster group pair. The order of cluster should be consistent with those in 'count_mat'.
pseudo_count	A pseudo count to add for counts in all cell types Default NULL means 0 except if a cell type is empty in one condition, otherwise pseudo_count will be: 0.01 * rowMeans for each condition
base_model	A string value: 'NULL' for 1 factor vs NULL factor testing; 'FULL' for FULL factors vs n-1 factors testing.
fix_phi	A numeric used to provided a fixed phi value for the GLM for all cell types
reference	A string value indicating which cell type is used as reference for normalization. 'NULL' indicates using total count for normalization.

Value

a list of significance p values for each cluster

Examples

K <- 3
totals1 = c(100, 800, 1300, 600)
totals2 = c(250, 700, 1100)
diri_s1 = rep(1, K) * 20
diri_s2 = rep(1, K) * 20
simil_mat = DCATS::create_simMat(K, confuse_rate=0.2)
sim_dat <- DCATS::simulator_base(totals1, totals2, diri_s1, diri_s2, simil_mat)
sim_count = rbind(sim_dat$numb_cond1, sim_dat$numb_cond2)
sim_design = data.frame(condition = c("g1", "g1", "g1", "g1", "g2", "g2", "g2"), 
gender = sample(c("Female", "Male"), 7, replace = TRUE))
## Using 1 factor vs NULL factor testing
dcats_GLM(sim_count, sim_design, similarity_mat = simil_mat)
## Using full factors vs n-1 factors testing with intercept term
dcats_GLM(sim_count, sim_design, similarity_mat = simil_mat, base_model='FULL')
## Fix phi
dcats_GLM(sim_count, sim_design, similarity_mat = simil_mat, fix_phi = 1/61)
## Specify reference cell type
colnames(sim_count) <- c("celltypeA", "celltypeB", "celltypeC")

huangyh09/DCATS documentation built on Nov. 25, 2022, 7:02 a.m.