In emosca-cnr/dmfind002: Network diffusion-based analysis of omics data

devtools::load_all(".")

Introduction

The R package dmfind implements the algorithms described in @Bersanelli2016. In this tutorial we will use igraph package for preparing the input data and plotting the results.

Installation

The package can be installed from within R:

library(devtools)
install_github("emosca-cnr/dmfind002", build_vignettes = TRUE)
library(dmfind)

Input data

The package requires an adjacency matrix and one or more input vectors. Let's create a simulated input dataset, including: a normalized adjacency matrix (A), a two-columns matrix (D) with two vectors representing descriptive statistics of two sample groups, and a binary vector where positive elements indicates the significant elements of an inferential statistics (u):

library(igraph)

edge_list_ig <- list(c(2, 3, 4, 12), c(1,3,4,5, 13), c(1, 2, 3), c(1,2,5), c(2,4, 7, 9), 7,
                              c(5,6,8), c(7,10), c(5,10), c(8,9,11), 10, 1, 2)
edge_list <- do.call(rbind, lapply(1:length(edge_list_ig),
                                   function(i) cbind(i, edge_list_ig[[i]])))
G <- graph_from_adj_list(edge_list_ig, 'all')
N <- length(V(G))
V(G)$name <- 1:N
A <- dmfind::normalize_adj_mat(as.matrix(as_adjacency_matrix(G)))
D <- matrix(abs(c(rnorm(N, 0, 0.2), rnorm(4, 2), rnorm(N-4, 0, 0.2))), ncol = 2,
            dimnames = list(1:N, 1:2))
temp_tgt <- D[1:4, 2]
temp_src_idx <- order(-D[, 2])[1:4]
temp_src <- D[temp_src_idx, 2]
D[1:4, 2] <- temp_src
D[temp_src_idx, 2] <- temp_tgt
u <- matrix(c(rep(1, 4), rep(0, N-4)), ncol=1, dimnames = list(1:N))
print(D)
print(u)

Note that elements ${1, 2, 3, 4}$ are enriched in information in D[, 2] in comparison to D[, 1], and are positive in $u$. At this point we can calculate the network smoothing index $(S)$ of each sample group and its variation $\Delta S$:

dS_Ds <- dmfind::calc_dS(D, A, classes = c(1, 2), eps = 4)
print(dS_Ds)
plot(dS_Ds$dX0, dS_Ds$dS, xlab='dX0', ylab='dS', pch='')
text(dS_Ds$dX0, dS_Ds$dS, rownames(dS_Ds), cex=0.7)

In case of the vector $u$ we calculate the permutation-adjusted network smoothing index $(Sp)$:

Sp_u <- dmfind::calc_Sp(u, A, k=9, eps = 4)
print(Sp_u)

Network resampling allows to assess the list ranked by $\Delta S$ (or $Sp$) for the presence of significantly connected components:

res <- dmfind::NR(edge_list, sort(array(dS_Ds$dS, dimnames = list(rownames(dS_Ds))), decreasing = TRUE), k=99)
print(res$NR_summary)
plot(log10(res$NR_summary$p), xlab='rank', type='l', ylab="log10(p)", ylim=c(-2,0))

Let's plot the network underlying the region with the highest $\Delta S$:

V(G)$color <- heat.colors(16)[
  dmfind::round(
    linear_map(
      res$NR_summary$ranking_score[match(V(G)$name, names(res$NR_summary$ranking_score))], 1, 16))]
par(mar=c(0, 0, 0, 0))
plot.igraph(G, vertex.size=10, vertex.label.cex=1)

Reference

---

references: - id: Bersanelli2016 title: Network diffusion-based analysis of high-throughput data for the detection of differentially enriched modules author: - family: Bersanelli given: Matteo - family: Mosca given: Ettore - family: Remondini given: Daniel - family: Castellani given: Gastone - family: Milanesi given: Luciano container-title: Scientific Reports volume: 6 DOI: 10.1038/srep34841 page: 34841 type: article-journal issued: year: 2016

---

emosca-cnr/dmfind002 documentation built on May 16, 2024, 10:44 p.m.