| Title: | Heatmaps for Multiple Network Data |
|---|---|
| Description: | Simplify the exploratory data analysis process for multiple network data sets with the help of hierarchical clustering, consensus clustering and heatmaps. Multiple network data consists of multiple disjoint networks that have common variables (e.g. ego networks). This package contains the necessary tools for exploring such data, from the data pre-processing stage to the creation of dynamic visualizations. |
| Authors: | Philippe Boileau [aut, cre] |
| Maintainer: | Philippe Boileau <[email protected]> |
| License: | MIT + file LICENSE |
| Version: | 2.1.1 |
| Built: | 2024-11-15 03:52:28 UTC |
| Source: | https://github.com/philboileau/neatmaps |
aggNodeAttr creates a data frame that summarizes node attributes.
aggNodeAttr(node_df, measure_of_cent = "mean")aggNodeAttr(node_df, measure_of_cent = "mean")
node_df |
A data frame containing all the characteristics of the nodes in the network. If there are n networks, a maximum of x nodes per network and y variables for each node, the data frame should have n rows and x*y columns. The column names of each variable should be written as follows: var1, var2, ... , varX. |
measure_of_cent |
A vector that contains the measures of centrality with which to summarize the node attributes. The supported measures are "mean" and "median". Note that missing values are excluded from the calculations. |
Philippe Boileau , [email protected]
consensusClusterNoPlots is a wrapper function for
ConsensusClusterPlus that suppresses
the creation of the plots that are created automatically.
calcICLNoPlots(consensus_results)calcICLNoPlots(consensus_results)
consensus_results |
Results of consensus clustering. The second item
in the list returned by |
Philippe Boileau , [email protected]
consClustResTable create a dataframe of the consensus cluster results.
The dataframe presents the results of each iteration of the
ConsensusClusterPlus algorithm, the
cluster consensus of each cluster and the list of the cluster elements with
their corresponding item consensus. The item consensus is taken with respect
to the variable's cluster allocation.
consClustResTable(neatmap_res)consClustResTable(neatmap_res)
neatmap_res |
Output from the |
A dataframe of the results of the consensus clustering.
Philippe Boileau , [email protected]
For more information on the consensus cluster and item consensus statistics, see Monti et al..
# create the data frame using the network, node and edge attributes df <- netsDataFrame(network_attr_df, node_attr_df, edge_df) # run the neatmap code on df neat_res <- neatmap(df, scale_df = "ecdf", max_k = 3, reps = 100, xlab = "vars", ylab = "nets", xlab_cex = 1, ylab_cex = 1) # get the consensus cluster results for each iteration consensus_res_df <- consClustResTable(neat_res)# create the data frame using the network, node and edge attributes df <- netsDataFrame(network_attr_df, node_attr_df, edge_df) # run the neatmap code on df neat_res <- neatmap(df, scale_df = "ecdf", max_k = 3, reps = 100, xlab = "vars", ylab = "nets", xlab_cex = 1, ylab_cex = 1) # get the consensus cluster results for each iteration consensus_res_df <- consClustResTable(neat_res)
consensusChangeECDF plots the relative change in area under empirical
cumulative distribution function for consecutive consensus cluster matrices
produced using the neatmap function.
consensusChangeECDF(neatmap_res)consensusChangeECDF(neatmap_res)
neatmap_res |
Output from the |
A ggplot of the change in consecutive area under the ECDFs of the consensus cluster matrices.
Philippe Boileau, [email protected]
For more information on the consensus matrices, see Monti et al..
#' # create the data frame using the network, node and edge attributes df <- netsDataFrame(network_attr_df, node_attr_df, edge_df) # run the neatmap code on df neat_res <- neatmap(df, scale_df = "ecdf", max_k = 3, reps = 100, xlab = "vars", ylab = "nets", xlab_cex = 1, ylab_cex = 1) # visualize the relative change in AU ECDF of consecutive consensus cluster # iterations consensusChangeECDF(neat_res)#' # create the data frame using the network, node and edge attributes df <- netsDataFrame(network_attr_df, node_attr_df, edge_df) # run the neatmap code on df neat_res <- neatmap(df, scale_df = "ecdf", max_k = 3, reps = 100, xlab = "vars", ylab = "nets", xlab_cex = 1, ylab_cex = 1) # visualize the relative change in AU ECDF of consecutive consensus cluster # iterations consensusChangeECDF(neat_res)
consensusClusterNoPlots is a wrapper function for
ConsensusClusterPlusthat suppresses
the creation of the plots that are created automatically.
consensusClusterNoPlots(df, link_method, dist_method, max_k, reps, p_var, p_net, cc_seed)consensusClusterNoPlots(df, link_method, dist_method, max_k, reps, p_var, p_net, cc_seed)
df |
A dataframe of network attributes containing only numeric values. The columns of the dataframe should likely be normalized. |
link_method |
The agglomeration method to be used for hierarchical
clustering. Defaults to the average linkage method. See other methods in
|
dist_method |
The distance measure to be used between columns and
between rows of the dataframe. Distance is used as a measure of similarity.
Defaults to euclidean distance. See other options in
|
max_k |
The maximum number of clusters to consider in the consensus clustering step. Consensus clustering will be performed for max_k-1 iterations, i.e. for 2, 3, ..., max_k clusters. Defaults to 10. |
reps |
The number of subsamples taken at each iteration of the consensus cluster algorithm. Defaults to 1000. |
p_var |
The proportion of network variables to be subsampled during consensus clustering. Defaults to 1. |
p_net |
The proportion of networks to be subsampled during consensus clustering. Defaults to 0.8. |
cc_seed |
The seed used to ensure the reproducibility of the consensus clustering. Defaults to 1. @author Philippe Boileau , [email protected] @importFrom ConsensusClusterPlus ConsensusClusterPlus @importFrom grDevices png dev.off |
consensusECDF plots the empirical cumulative distribution functions
(ECDF) of the consensus matrices produced during the consensus clustering
step of the neatmap function.
consensusECDF(neatmap_res)consensusECDF(neatmap_res)
neatmap_res |
Output from the |
This function visualizes the ECDFs of the consensus matrices for each each
iteration of consensus clustering that is carried out as part of the
neatmap function.
Returns a ggplot depicting the ECDFs of each iteration of the consensus clustering, i.e. one ECDF per number of clusters used in each iteration.
Philippe Boileau , [email protected]
For more information on the consensus matrices, see Monti et al..
#' # create the data frame using the network, node and edge attributes df <- netsDataFrame(network_attr_df, node_attr_df, edge_df) # run the neatmap code on df neat_res <- neatmap(df, scale_df = "ecdf", max_k = 3, reps = 100, xlab = "vars", ylab = "nets", xlab_cex = 1, ylab_cex = 1) # create the ECDF plot consensusECDF(neat_res)#' # create the data frame using the network, node and edge attributes df <- netsDataFrame(network_attr_df, node_attr_df, edge_df) # run the neatmap code on df neat_res <- neatmap(df, scale_df = "ecdf", max_k = 3, reps = 100, xlab = "vars", ylab = "nets", xlab_cex = 1, ylab_cex = 1) # create the ECDF plot consensusECDF(neat_res)
consensusMap produces a list of heatmaps from the consensus matrices
produced during the consensus clustering step of the neatmap
function.
consensusMap(neatmap_res, link_method = "average")consensusMap(neatmap_res, link_method = "average")
neatmap_res |
Output from the |
link_method |
The agglomeration method to be used for hierarchical
clustering. Defaults to the average linkage method. See other methods in
|
This function will create a list of heatmaps of the consensus matrices
produced during the consensus clustering step of the neatmap
function. The default clustering method used in the heatmaps is hierarchical
clustering using the average linkage method, though other linkage methods
can be used. The consensus cluster matrix is used as a measure of similarity.
The heatmaps are produced using heatmaply.
Returns of a list of heatmaps depicting the consensus matrices of each
Philippe Boileau, [email protected]
For more information on the consensus matrices, see Monti et al..
# create the data frame using the network, node and edge attributes df <- netsDataFrame(network_attr_df, node_attr_df, edge_df) # run the neatmap code on df neat_res <- neatmap(df, scale_df = "ecdf", max_k = 3, reps = 100, xlab = "vars", ylab = "nets", xlab_cex = 1, ylab_cex = 1) # create the list of heatmaps for each iteration hm_list <- consensusMap(neat_res)# create the data frame using the network, node and edge attributes df <- netsDataFrame(network_attr_df, node_attr_df, edge_df) # run the neatmap code on df neat_res <- neatmap(df, scale_df = "ecdf", max_k = 3, reps = 100, xlab = "vars", ylab = "nets", xlab_cex = 1, ylab_cex = 1) # create the list of heatmaps for each iteration hm_list <- consensusMap(neat_res)
createNetworks creates Igraph network objects using an edge data
frame. This is important for computing structural properties of the
networks to be explored by neatmap.
createNetworks(edge_df)createNetworks(edge_df)
edge_df |
A data frame where each row represents a different network and where each column represents a potential edge between node A and node B. The column names should be of the form "XA_B", where A and B are the node numbers in the network. If Node A or B do not exist in the specific network, the cell should have a value of NA. If there is no edge between A and B, place a value of 0. Avoid redundant column names since all edges are assumed to be undirected, e.g. avoid "XA_B" and "XB_A". |
Philippe Boileau, [email protected]
A dataset containing a list of undirected edges for ten different networks. Each network has a maximum size of 5 nodes. The network and node attribute data are saved in their respective files.
edge_dfedge_df
An object of class data.frame with 10 rows and 10 columns.
Philippe Boileau, [email protected]
getStructureAttr produces a data frame of the structural attributes
of a list of networks.
getStructureAttr(net_list)getStructureAttr(net_list)
net_list |
A list of Igraph network objects that represent the collection of networks. |
Philippe Boileau, [email protected]
hierarchy calculates the hierarchy of a network
hierarchy(net)hierarchy(net)
net |
An igraph object representing a network |
Philippe Boileau, [email protected]
neatmap produces a heatmap of multi-network data and identifies stable
clusters in its variables.
neatmap(df, scale_df, link_method = "average", dist_method = "euclidean", max_k = 10, reps = 1000, p_var = 1, p_net = 0.8, cc_seed = 100, main_title = "", xlab, ylab, xlab_cex = 1, ylab_cex = 1, heatmap_margins = c(50, 50, 50, 100))neatmap(df, scale_df, link_method = "average", dist_method = "euclidean", max_k = 10, reps = 1000, p_var = 1, p_net = 0.8, cc_seed = 100, main_title = "", xlab, ylab, xlab_cex = 1, ylab_cex = 1, heatmap_margins = c(50, 50, 50, 100))
df |
a dataframe of network attributes containing only numeric values. |
scale_df |
A string indicating whether the columns of the data frame should be scaled, and, if so, which method should be used. The options are "none", "ecdf", "normalize" and "percentize". If "none" is selected, then the columns are not scaled. If "ecdf" is selected, then the columns are transformed into their empirical cumulative distribution. If "normalize" is selected, each column is centered to have a mean of 0 and scaled to have a standard deviation of 1. If "percentize" is selected, column values are transformed into percentiles. |
link_method |
The agglomeration method to be used for hierarchical
clustering. Defaults to the average linkage method. See other methods in
|
dist_method |
The distance measure to be used between columns and
between rows of the dataframe. Distance is used as a measure of similarity.
Defaults to euclidean distance. See other options in
|
max_k |
The maximum number of clusters to consider in the consensus clustering step. Consensus clustering will be performed for max_k-1 iterations, i.e. for 2, 3, ..., max_k clusters. Defaults to 10. |
reps |
The number of subsamples taken at each iteration of the consensus cluster algorithm. Defaults to 1000. |
p_var |
The proportion of network variables to be subsampled during consensus clustering. Defaults to 1. |
p_net |
The proportion of networks to be subsampled during consensus clustering. Defaults to 0.8. |
cc_seed |
The seed used to ensure the reproducibility of the consensus clustering. Defaults to 1. |
main_title |
The title of the heatmap. |
xlab |
The x axis label of the heatmap. |
ylab |
The y axis label of the heatmap. |
xlab_cex |
The font size of the elements on the x axis. |
ylab_cex |
The font size of the elements on the y axis. |
heatmap_margins |
The size of the margins for the heatmap.
See |
This function allows users to efficiently explore their multi-network data
by visualizing their data with a heatmap and assessing the stability of the
associations presented within it. neatmap requires that the data
frame be processed into an appropriate format prior to use. Data is then
scaled (if necessary) using of the built in methods. See (list functions) for
further details on how to prepare multi-network data for use with
neatmap. The heatmap is created using
heatmaply and the consensus clustering is performed
using ConsensusClusterPlus
A named list containing the heatmap of the multi-network data and a
list of length max_k-1 where each element is a list containing the
consensus matrix, the consensus hierarchical clustering results and the
consensus class assignments. The list of results produced by the consensus
clustering can be parsed using following functions in the
neatmaps package: consClustResTable,
consensusECDF and consensusChangeECDF.
Philippe Boileau, [email protected]
For more information on the consensus clustering, see Monti et al..
# create the data frame using the network, node and edge attributes df <- netsDataFrame(network_attr_df, node_attr_df, edge_df) # run the neatmap code on df neat_res <- neatmap(df, scale_df = "ecdf", max_k = 3, reps = 100, xlab = "vars", ylab = "nets", xlab_cex = 1, ylab_cex = 1) # extract the heatmap heatmap <- neat_res$heatmap # extract the consensus clustering results consensus_res <- neat_res$consensus_clust# create the data frame using the network, node and edge attributes df <- netsDataFrame(network_attr_df, node_attr_df, edge_df) # run the neatmap code on df neat_res <- neatmap(df, scale_df = "ecdf", max_k = 3, reps = 100, xlab = "vars", ylab = "nets", xlab_cex = 1, ylab_cex = 1) # extract the heatmap heatmap <- neat_res$heatmap # extract the consensus clustering results consensus_res <- neat_res$consensus_clust
neatmaps packageA package for exploring multi-network data.
See the README on CRAN or GitHub
Philippe Boileau, [email protected]
netsDataFrame produces data frames of collections of networks.
netsDataFrame(net_attr_df, node_attr_df, edge_df, cent_measure = c("mean"))netsDataFrame(net_attr_df, node_attr_df, edge_df, cent_measure = c("mean"))
net_attr_df |
A data frame consisting of all of the networks' graph attributes. The first column should contain the name of the network, and all other columns should be numeric. All empty entries should be filled as "NA". |
node_attr_df |
A data frame consisting of all of the networks' nodes' attributes. All columns should be numeric. All empty entries should be filled in as "NA". |
edge_df |
A data frame consisting of the edge matrix for each ego network. Edges are assumed to be undirected and unweighted. 1 indicates the existence of an edge between nodes, 0 indicates the lack of an edge. |
cent_measure |
A vector of the measures of centrality to be used for the summary of the node attributes data. The supported measures of centrality are: "mean" and "median". |
The function produces data frames of collections of networks. The function requires the input of three data frames: a data frame containing the graph attributes, a data frame containing the node characteristics and a data frame containing the edge list of each network. The rows in each of these data frames must represent individual networks, and must therefore have identical row length. Measures of centrality used in the summarization of the node attributes must also be furnished.
The function returns a data frame that offers an overview of all of the ego networks.
Philippe Boileau, [email protected]
df <- netsDataFrame(network_attr_df, node_attr_df, edge_df)df <- netsDataFrame(network_attr_df, node_attr_df, edge_df)
A data set containing four randomly generated variables used to mimic the network attributes of ten different networks. Attribute 1 and 2 have a correlation of 0.41, 1 and 3 have a correlation of 0.91, 1 and 4 have a correlation of 0.34, 2 and 3 have a correlation of 0.07, 2 and 4 have a correlation of 0.17 and 3 and 4 have a correlation of 0.32.
network_attr_dfnetwork_attr_df
An object of class data.frame with 10 rows and 4 columns.
Philippe Boileau, [email protected]
A dataset containing randomly generated node attributes for each node in each of the ten networks. Attributes A and B have a correlation of roughly 0.8, A and C have a correlation of roughly -0.2, B and C have a correlation of roughly 0.5. Attributes D and E were generated completely randomly, and should not be strongly correlated with any of the other attributes.
node_attr_dfnode_attr_df
An object of class data.frame with 10 rows and 25 columns.
Philippe Boileau, [email protected]
scaleColumns scales the columns of a data frame object between the
values of 0 and 1 without changing the underlying distribution of the
columns.
scaleColumns(df)scaleColumns(df)
df |
The data frame of numerical values to be scaled. |
Philippe Boileau, [email protected]