SEMgraph (version 1.2.2)

SEMdci: SEM-based differential network analysis

Description

Creates a sub-network with perturbed edges obtained from the output of SEMace, comparable to the procedure in Jablonski et al (2022), or of SEMrun with two-group and CGGM solver, comparable to the algorithm 2 in Belyaeva et al (2021). To increase the efficiency of computations for large graphs, users can select to break the network structure into clusters, and select the topological clustering method (see clusterGraph). The function SEMrun is applied iteratively on each cluster (with size min > 10 and max < 500) to obtain the graph with the full list of perturbed edges.

Usage

SEMdci(graph, data, group, type = "ace", method = "BH", alpha = 0.05, ...)

Value

An igraph object.

Arguments

graph

Input network as an igraph object.

data

A matrix or data.frame. Rows correspond to subjects, and columns to graph nodes (variables).

group

A binary vector. This vector must be as long as the number of subjects. Each vector element must be 1 for cases and 0 for control subjects.

type

Average Causal Effect (ACE) with two-group, "parents" (back-door) adjustement set, and "direct" effects (type = "ace", default), or CGGM solver with two-group using a clustering method. If type = "tahc", network modules are generated using the tree agglomerative hierarchical clustering method, or non-tree clustering methods from igraph package, i.e., type = "wtc" (walktrap community structure with short random walks), type ="ebc" (edge betweeness clustering), type = "fgc" (fast greedy method), type = "lbc" (label propagation method), type = "lec" (leading eigenvector method), type = "loc" (multi-level optimization), type = "opc" (optimal community structure), type = "sgc" (spinglass statistical mechanics), type = "none" (no breaking network structure into clusters).

method

Multiple testing correction method. One of the values available in p.adjust. By default, method is set to "BH" (i.e., FDR multiple test correction).

alpha

Significance level (default = 0.05) for edge set selection.

...

Currently ignored.

Author

Mario Grassi

References

Belyaeva A, Squires C, Uhler C (2021). DCI: learning causal differences between gene regulatory networks. Bioinformatics, 37(18): 3067–3069. <https://doi: 10.1093/bioinformatics/btab167>

Jablonski K, Pirkl M, Ćevid D, Bühlmann P, Beerenwinkel N (2022). Identifying cancer pathway dysregulations using differential causal effects. Bioinformatics, 38(6):1550–1559. <https://doi.org/10.1093/bioinformatics/btab847>

Examples

if (FALSE) {

#load SEMdata package for ALS data with 17K genes:
#devtools::install_github("fernandoPalluzzi/SEMdata")
#library(SEMdata)

# Nonparanormal(npn) transformation
library(huge)
data.npn<- huge.npn(alsData$exprs)
dim(data.npn) #160 17695

# Extract KEGG interactome (max component)
KEGG<- properties(kegg)[[1]]
summary(KEGG)

# KEGG modules with ALS perturbed edges using fast gready clustering
gD<- SEMdci(KEGG, data.npn, alsData$group, type="fgc")
summary(gD)
gcD<- properties(gD)

old.par <- par(no.readonly = TRUE)
par(mfrow=c(2,2), mar=rep(2,4))
gplot(gcD[[1]], l="fdp", main="max component")
gplot(gcD[[2]], l="fdp", main="2nd component")
gplot(gcD[[3]], l="fdp", main="3rd component")
gplot(gcD[[4]], l="fdp", main="4th component")
par(old.par)

}