Simulate multi-platform data from multiple Gaussian graphical models

Generates S platforms (data types) of data, each with the same K groups, with graphs that share structure across platforms and across groups.

Usage

simulate_multiggm_platform(
  K = 2,
  S = 2,
  p_vec = c(20, 15),
  n = 100,
  graph_type = c("band", "random", "hub"),
  edge_prob = 0.1,
  perturb_prob = 0.1,
  platform_perturb_prob = 0.2,
  signal = c(0.3, 0.6),
  seed = NULL
)

Arguments

K

Integer; number of sample groups. Default 2.

S

Integer; number of platforms (data types). Default 2.

p_vec

Integer vector of length S; number of variables per platform. Default c(20, 15).

n

Integer (scalar or length-K vector); sample size per group (same across platforms). Default 100.

graph_type

Character; type of base graph. See simulate_multiggm.

edge_prob

Numeric; edge probability of the base graph for graph_type = "random". Default 0.1.

perturb_prob

Numeric; probability of edge flip between groups within a platform. Default 0.1.

platform_perturb_prob

Numeric; probability of edge flip between platform 1 and each other platform. Default 0.2.

signal

Numeric vector of length 2; magnitude range for off-diagonal precision entries. Default c(0.3, 0.6).

seed

Optional integer random seed.

Value

A list with components:

• platform_data: List of S elements, each suitable as input to multiggm_mcmc(method = "ssvs_platform", platform_data = ...). Each element is a list with data_list, S_list, n_vec, Omega_list, and adj_list.

• adj_base: The base graph on max(p_vec) nodes.

• platform_adj: List of S platform graphs (before group perturbation).

• K, S, p_vec: Dimensions.

Details

The data-generating process is hierarchical:

1. One base graph is drawn on max(p_vec) nodes (see graph_type).

2. Platform s uses the leading p_vec[s] x p_vec[s] block of the base graph. Platform 1 uses it unchanged; platforms 2, ..., S flip each pair independently with probability platform_perturb_prob. Variable j therefore plays the same role in every platform that has it.

3. Within each platform, group 1 uses the platform graph and groups 2, ..., K flip each pair with probability perturb_prob, exactly as in simulate_multiggm(). Group perturbations are independent across platforms.

4. Precision matrices and data are generated from each graph as in simulate_multiggm().

See also

simulate_multiggm(), multiggm_mcmc()

Examples

sim <- simulate_multiggm_platform(K = 2, S = 2, p_vec = c(10, 8),
                                   n = 80, seed = 42)
str(sim, max.level = 2)
#> List of 6
#>  $ platform_data:List of 2
#>   ..$ :List of 5
#>   ..$ :List of 5
#>  $ adj_base     : int [1:10, 1:10] 0 1 1 0 0 0 0 0 0 0 ...
#>  $ platform_adj :List of 2
#>   ..$ : int [1:10, 1:10] 0 1 1 0 0 0 0 0 0 0 ...
#>   ..$ : int [1:8, 1:8] 0 0 1 0 0 0 0 0 0 0 ...
#>  $ K            : num 2
#>  $ S            : num 2
#>  $ p_vec        : int [1:2] 10 8