neatmaps

Overview

The goal of the neatmaps package is to simplify the exploratory data analysis process for multiple network data sets with the help of heatmaps and consensus clustering. Multiple network data consists of multiple disjoint networks that share common variables. Ego networks are an example of such data sets. This package contains tools necessary to prepare raw multiple network data for analysis, create a heatmap of the data, perform consensus clustering on the networks’ variables and assess the stability of associations among variables depicted in the heatmap.

Installation

# To install neatmaps, simply run the following code:
install.packages('neatmaps')

Code Example

Below is an example of how to use the key functions in the neatmaps package on a simulated network dataset. Run this code locally to produce the plots.

First, load the package and format the network data using the netsDataFrame function. This function has four inputs: an nxp data frame of n networks described by p graph attributes (e.g. density), an *nxm*k data frame of n networks, their m nodes and k variables describing these nodes (e.g. age of node, in a social network context), a data frame containing the undirected adjacensy matrix and a string defining the aggregation function used to summarize the node variable data.

The data frame is then passed to the neatmap function. The data is scaled using the ecdf method, and the maximum number of clusters among the aggregated variables to be considered by consensus clusterting is set to five. For each iteration of consensus clustering, 1000 bootstrap repititons are computed.

library(neatmaps)

set.seed(215)

df <- netsDataFrame(network_attr_df,
                    node_attr_df,
                    edge_df)

neat_res <- neatmap(df, scale_df = "ecdf", max_k = 5, reps = 1000, 
                    xlab = "vars", ylab = "nets", xlab_cex = 1, ylab_cex = 1)

Next, the heatmap is plotted.

neat_res$heatmap

Finally, the results of the consensus clustering are visualized to identify the stable clusters of variables in the heatmap. The consensus matrices are presented first, followed by the ECDFs of the consensus matrices and finally the relative change in ECDF of consecutive iterations of the consensus clustering algorithm.

consensusMap(neat_res)

consensusECDF(neat_res)

consensusChangeECDF(neat_res)

These results indicate that there are likely four distinct clusters among the aggregated network variables, since the consensus matrix with four clusters offers the shapest contrasts between clusters, it’s ECDF most closely resembles that of an indicator function, and because of the elbow in the relative change of the ECDF occurs when there are four clusters.

Further documentation is available on CRAN.