ICMg.combined.sampler {ICMg}R Documentation

ICMg: inferring gene modules from combinations of interaction and expression data.

Description

ICMg.combined.sampler computes samples from the posterior of the assignments of datapoints (interactions and expression profiles) to latent components. From these we can then obtain component membership distributions and clusterings for genes.

Usage

ICMg.combined.sampler(L, X, C, alpha=10, beta=0.01, pm0=0, V0=1, V=0.1, B.num=8, B.size=100, S.num=20, S.size=10, C.boost=1)

Arguments

L N x 2 matrix of link endpoints (N = number of links).
X M x D matrix of gene expression profiles (M = number of nodes, D = number of observations).
C Number of components.
alpha Hyperparameter describing the global distribution over components, larger alpha gives a more uniform distribution.
beta Hyperparameter describing the component-wise distributions over nodes, larger beta gives a more uniform distribution.
pm0 Hyperparameter describing the prior mean of the expression profiles, should be zero.
V0 Hyperparameter describing the variation of the component-wise expression profiles means around pm0.
V Hyperparameter describing the variation of gene-specific expression profiles around the component-wise means.
B.num Number of burnin rounds.*
B.size Size of one burnin round.*
S.num Number of sample rounds.*
S.size Size of one sample round.*
C.boost Set to 1 to use faster iteration with C, set to 0 to use slower R functions.

Details

* One run consists of two parts, during burnin the sampler is expected to mix, after which the samples are taken. Information about convergence (convN and convL are estimates of convergence for link and node sampling, respectively) and component sizes are printed after each burnin/sample round. For example: B.num=8, B.size=100, S.num=20, S.size=10, runs 800 burnin iterations in 8 rounds and then takes 20 samples with an interval of 10 iterations.

Value

Returns samples as a list:

z S.num x N matrix of samples of component assignments for links.
w S.num x M matrix of samples of component assignments for gene expression profiles.
convl Vector of length (B.num + S.num) with convergence estimator values for link sampling.
convn Vector of length (B.num + S.num) with convergence estimator values for node sampling.
countsl (B.num + S.num) x C matrix of link component sizes.
countsn (B.num + S.num) x C matrix of node component sizes.

additionally all parameters of the run are included in the list.

Author(s)

Juuso Parkkinen

See Also

ICMg.links.sampler

Examples

        ## Load data and set parameters
        data(osmo)
        C.boost = 1 ## Use faster C funtions
        alpha = 10
        beta = 0.01
        B.num = 10
        B.size = 10
        S.num = 10  
        S.size = 10
        C = 24
        pm0 = 0
        V0 = 1               
        V = 0.1

        ## Run sampling
        res = ICMg.combined.sampler(osmo$ppi, osmo$exp, C, alpha, beta, pm0, V0, V, B.num, B.size, S.num, S.size, C.boost) 
        ## Compute component membership probabilities for nodes
        res$comp.memb <- ICMg.get.comp.memberships(osmo$ppi, res)
        ## Compute (hard) clustering for nodes
        res$clustering <- apply(res$comp.memb, 2, which.max)
[Package ICMg version 1.0 Index]