simulate_multicluster: Simulate multicluster counts with time dependent association...

In retogerber/censcyt: Differential abundance analysis with a right censored covariate in high-dimensional cytometry

Simulate multicluster counts with time dependent association from a Dirichlet-Multinomial distribution

Description

Simulate multicluster counts with time dependent association from a Dirichlet-Multinomial distribution

Usage

simulate_multicluster(
  counts = NULL,
  nr_diff = 2,
  nr_samples = NULL,
  alphas = NULL,
  theta = NULL,
  sizes = NULL,
  covariate = NULL,
  slope = NULL,
  group = NULL,
  group_slope = NULL,
  diff_cluster = FALSE,
  enforce_sum_alpha = FALSE,
  return_summarized_experiment = FALSE
)

Arguments

counts	the reference counts data set, either a matrix with rows as cluster and colums as samples or a SummarizedExperiment-class object as generated from calcCounts.
nr_diff	number of clusters where an association should be introduced. Has to be an even number.
nr_samples	number of samples in output data. If NULL will set to same as input data.
alphas	alpha parameter of Dirichlet-Multinomial distribution. If 'NULL' will be estimated from 'counts'.
theta	correlation parameter. If 'NULL' will be estimated from 'counts'.
sizes	total sizes for each sample
covariate	covariates, one for each sample. Default Null means random draws from an exponential distribution with rate = 1.
slope	negative double. Coefficients corresponding to the covariate for the DA clusters. One for each pair of DA clusters. To ensure correctness of the final distribution use only negative values. Alternatively can be a list of length 'nr_diff'/2, where each elements indicates the proportion of the cluster size at the maximum covariate relative to the mean. E.g. 0.1 means that the cluster proportion at the maximum covariate is 0.1 times smaller than the mean.
group	either Null (no group effect), double between 0 and 1 (proportion of samples with group effect), integer (total number of samples with group effect), vector of 0 and 1 (indicating which samples have a group effect) or TRUE (effect with even group size).
group_slope	regression coefficient of second covariate 'group'. If Null will choose a value automatically. Alternatively can be a list of length 'nr_diff'/2, where each elements indicates the proportion of the cluster size at the maximum covariate relative to the mean. E.g. 0.1 means that the cluster proportion at the maximum covariate is 0.1 times smaller than the mean.
diff_cluster	Logical. Should the clusters be choosen random (TRUE) or according to a minimal distance of of mean cluster sizes (FALSE). Alternatively a list of length 'nr_diff' with each element a vector of length 2 indicating the paired clusters can be given. Default is FALSE.
enforce_sum_alpha	Logical. Should the total sum of alphas be kept constant to ensure randomness of non association clusters. The drawback is that one of the two paired clusters with an association will not follow a GLMM (binomial link function) exactly any more. Default is TRUE.
return_summarized_experiment	logical. Should the counts returned as a SummarizedExperiment-class object. Default is FALSE.

Value

returns a list with elements counts (either matrix or SummarizedExperiment object, depending on input), row_data (data per cluster: regression coefficients used), col_data (data per sample: covariates), alphas (matrix of alpha parameters used), theta (theta parameter), var_counts (covariance matrix of a DM distribution with the given alphas and sizes).

Examples

# without data reference:
alphas <- runif(20,10,100)
sizes <- runif(10,1e4,1e5)
output <- simulate_multicluster(alphas=alphas,sizes=sizes)
# counts:
counts <- output$counts

# with data reference:
# first simulate reference data set (normally this would be a real data set):
data <- t(dirmult::simPop(n=runif(10,1e4,1e5),theta=0.001)$data)
# then generate new data set based on original one but if DA clusters
output <- simulate_multicluster(data)

# specify number of differential clusters (has to be an even number):
output <- simulate_multicluster(alphas=alphas,sizes=sizes,nr_diff = 4)

# specify which clusters should be differential:
output <- simulate_multicluster(alphas=alphas,
                                sizes=sizes,
                                nr_diff = 4, 
                                diff_cluster = list(c(2,9),c(6,7)))

# with second covariate (group):
output <- simulate_multicluster(alphas=alphas,sizes=sizes, group = TRUE)

# with second covariate (group), specify group proportion:
output <- simulate_multicluster(alphas=alphas,sizes=sizes, group = 0.5)

# with second covariate (group), specify id of group memberships for one group:
output <- simulate_multicluster(alphas=alphas,sizes=sizes, group = 3:7)

retogerber/censcyt documentation built on Feb. 7, 2023, 9:56 a.m.