NIMAA: A package for Nominal Data Mining Analysis.

Description

The NIMAA package contains 13 main functions that together provide a pipeline for nominal data mining. This package also includes four datasets that can be used to perform various nominal data mining analyses.

General properties of the network

Description

Generic function for network properties, allowing you to get a quick overview of the network topology.

Usage

analyseNetwork(graph)

Arguments

Details

The following measurements are calculated in one step using the igraph package to analyze the input graph object: the Degree, Betweenness, Closeness, Kleinberg's hub score and Eigenvector centrality of nodes (vertices), the Betweenness centrality of edges, number of nodes, edges and components, the edge density, global Eigenvector centrality value and global Kleinberg's hub centrality score.

Value

A list containing vertices and edges centrality values as well as general_stats for the whole network.

See Also

vcount, ecount, edge_density, count_components, degree,betweenness, edge_betweenness,closeness, eigen_centrality,hub_score.

Examples

# generate a toy graph
g1 <- igraph::make_graph(c(1, 2, 3, 4, 1, 3), directed = FALSE)
igraph::V(g1)$name <- c("n1", "n2", "n3", "n4")

# generate random graph according to the Erdos-Renyi model
g2 <- igraph::sample_gnm(10, 23)
igraph::V(g2)$name <- letters[1:10]

# run analyseNetwork
analyseNetwork(g1)
analyseNetwork(g2)

beatAML

Description

The Beat AML program produced a comprehensive dataset on acute myeloid leukemia (AML) that included genomic data (i.e., whole-exome sequencing and RNA sequencing), ex vivo drug response, and clinical data. We have included only the drug response data from this program in this package.

Usage

beatAML

Format

A tibble with three variables:

inhibitor

names of inhibitors

patient_id

anonymous patient ID

median

the median of drug response at corresponding inhibitor and patient_id

Source

Tyner, Jeffrey W., et al.(2018), Functional Genomic Landscape of Acute Myeloid Leukaemia. Nature 562 (7728): 526–31. http://vizome.org/aml/

drugComb

Description

DrugComb is a web-based portal for storing and analyzing drug combination screening datasets, containing 739,964 drug combination experiments across 2320 cancer cell lines and 8397 drugs. Single drug screening was extracted and provided here as a subset of drug combination experiments.

Usage

drugComb

Format

A tibble with three variables:

inhibitor

names of inhibitors

cell_line

cell_line for each inhibitor

median

the median of drug response at corresponding inhibitor and cell_line

Source

Zheng, S., Aldahdooh, J., Shadbahr, T., Wang, Y., Aldahdooh, D., Bao, J., Wang, W., Tang, J. (2021). DrugComb update: a more comprehensive drug sensitivity data repository and analysis portal, Nucleic Acids Research, 49(W1), W174–W184. https://academic.oup.com/nar/article/49/W1/W174/6290546

Extract the non-missing submatrices from a given matrix.

Description

This function arranges the input matrix and extracts the submatrices with non-missing values or with a specific proportion of missing values (except for the elements-max submatrix). The result is also shown as plotly figure.

Usage

extractSubMatrix(
  x,
  shape = "All",
  verbose = FALSE,
  palette = "Greys",
  row.vars = NULL,
  col.vars = NULL,
  bar = 1,
  plot_weight = FALSE,
  print_skim = FALSE
)

Arguments

Details

This function performs row- and column-wise preprocessing in order to extract the largest submtrices. The distinction is that the first employs the original input matrix (row-wise), whereas the second employs the transposed matrix (column-wise). Following that, this function performs a "three-step arrangement" on the matrix, the first step being row-by-row arrangement, the second step being column-by-column arrangement, and the third step being total rearranging. Then, using four strategies, namely "Square", "Rectangular row", "Rectangular col", and "Rectangular element max", this function finds the largest possible submatrix (with no missing values), outputs the result, and prints the visualization. "Square" denotes the square submatric with the same number of rows and columns. "Rectangular_row" indicates the submatrices with the most rows. "Rectangular_col" denotes the submatrices with the most columns. "Rectangular_element_max" indicates the submatrices with the most elements which is typically a rectangular submatrix.

Value

A matrix or a list of matrices with non-missing (bar = 1) or a few missing values inside. Also, a specific heat map plot is generated to visualize the topology of missing values and the submatrix sub-setting from the original incidence matrix. Additionally, the nestedness temperature is included to indicate whether the original incidence matrix should be divided into several incidence matrices beforehand.

See Also

arrange, arrange_if

Examples

# load part of the beatAML data
beatAML_data <- NIMAA::beatAML[1:10000,]

# convert to incidence matrix
beatAML_incidence_matrix <- nominalAsBinet(beatAML_data)

# extract submatrices with non-missing values
sub_matrices <- extractSubMatrix(beatAML_incidence_matrix, col.vars = "patient_id",
 row.vars = "inhibitor")

Find clusters in projected unipartite networks

Description

This function looks for the clusters in the projected unipartite networks of the bipartite network (the incidence matrix) that was given to it.

Usage

findCluster(
  inc_mat,
  part = 1,
  method = "all",
  normalization = TRUE,
  rm_weak_edges = TRUE,
  rm_method = "delete",
  threshold = "median",
  set_remaining_to_1 = TRUE,
  extra_feature = NULL,
  comparison = TRUE
)

Arguments

Details

This function performs optional preprocessing, such as normalization, on the input incidence matrix (bipartite network). The matrix is then used to perform bipartite network projection and optional preprocessing on one of the projected networks specified, such as removing edges with low weights (weak edges). Additionally, the user can specify the removal method, threshold value, or binarization of the weights. For the networks obtained after processing, this function implements some clustering methods in igraph such as "walktrap" and "infomap", to detect the communities within the network. Furthermore, if external features (prior knowledge) are provided, the function compares the clustering results obtained with the external features in terms of similarity as an external validation of clustering. Otherwise, several internal validation criteria such as modularity and coverage are only represented to compare the clustering results.

Value

A list containing the igraph object of the projected network, the clustering results of each method on the projected network separately, along with a comparison between them. The applied clustering arguments and the network's distance matrix are also included in this list for potential use in the next steps. In the case of weighted projected networks, the distance matrix is obtained by inverting the edge weights. The comparison of selected clustering methods is also presented as bar plots simultaneously.

Examples

# generate an incidence matrix
data <- matrix(c(1, 0, 0, 0, 1, 1, 1, 0, 0, 0, 1, 1, 1, 0, 0), nrow = 3)
colnames(data) <- letters[1:5]
rownames(data) <- LETTERS[1:3]

# run findCluster() to do clustering
cls <- findCluster(
  data,
  part = 1,
  method = "all",
  normalization = FALSE,
  rm_weak_edges = TRUE,
  comparison = TRUE
)

herbIngredient

Description

The second version of the TCMID database was used as one of the large datasets in the TCM herb field. TCMID includes additional ingredient-specific experimental data based on herbal mass spectrometry spectra. The herb and ingredient associations from this database are provided here.

Usage

herbIngredient

Format

A tibble with two variables:

herb

Scientific Latin names of available herbs in TCM database

pubchem_id

unique compound id (CID) of ingredients based on PubChem database for available herb in TCMID database

Source

Huang, L., Xie, D., Yu, Y., Liu, H., Shi, Y., Shi, T., & Wen, C. (2018). TCMID 2.0: a comprehensive resource for TCM. Nucleic acids research, 46(D1), D1117–D1120. https://academic.oup.com/nar/article/46/D1/D1117/4584630

Convert nominal data to a bipartite network

Description

This function converts nominal data, which is represented in a data frame format as an edge list, to a bipartite network in the incidence matrix format. Nominal data is typically a data frame with two (or three) columns representing two nominal variables labels that NIMAA considers as starting and ending nodes (labels) in an edge list (or with a numeric value for each pairwise relationship of labels). The elements in the incidence matrix are the binary or numeric values of the pairwise relationships.

Usage

nominalAsBinet(
  el,
  index_nominal = c(1, 2),
  index_numeric = 3,
  print_skim = FALSE
)

Arguments

Value

The incidence matrix representing the corresponding bipartite network. If print_skim set to TRUE, a summary of the matrix is also provided.

See Also

pivot_wider, column_to_rownames

Examples

# generate a data frame with two nominal variables without numeric values
el1 <- data.frame(
  nominal_var_1 = c("d", "e", "e", "b", "a", "a"),
  nominal_var_2 = c("J", "N", "O", "R", "R", "L")
)

# generate a data frame with two nominal variables with numeric values
el2 <- data.frame(
  nominal_var_1 = c("d", "e", "e", "b", "a", "a"),
  nominal_var_2 = c("J", "N", "O", "R", "R", "L"),
  numeric_val = c(4, 5, 5, 8, 7, 7)
)

# run nominalAsBinet() to convert the edge list to the incidence matrix
inc_mat1 <- nominalAsBinet(el1)
inc_mat2 <- nominalAsBinet(el2)

Plot the bipartite network and the corresponding projected networks

Description

This function converts the incidence matrix into an igraph object and provides network visualization in multiple ways.

Usage

plotBipartite(
  inc_mat,
  part = 0,
  verbose = FALSE,
  vertex.label.display = FALSE,
  layout = layout.bipartite,
  vertex.shape = c("square", "circle"),
  vertix.color = c("steel blue", "orange"),
  vertex.label.cex = 0.3,
  vertex.size = 4,
  edge.width = 0.4,
  edge.color = "pink"
)

Arguments

Value

An igraph network object with visualization.

Examples

# load part of the beatAML data
beatAML_data <- NIMAA::beatAML[1:10000,]

# convert to incidence matrix
beatAML_incidence_matrix <- nominalAsBinet(beatAML_data)

# plot with the vertex label showing
plotBipartite(inc_mat = beatAML_incidence_matrix, vertex.label.display = TRUE)

Use the incidence matrix to plot an interactive bipartite network

Description

This function converts the input incidence matrix into a bipartite network, and generates a customized interactive bipartite network visualization.

Usage

plotBipartiteInteractive(inc_mat)

Arguments

Details

This function creates customized interactive visualization for bipartite networks. The user can enter a simple incidence matrix to generate a dynamic network with a bipartite network layout, in which two parts use different colors and shapes to represent nodes. This function relies on the visNetwork package.

Value

An visNetwork object for interactive figure.

See Also

visNetwork

Examples

# load part of the beatAML data
beatAML_data <- NIMAA::beatAML[1:10000,]

# convert to incidence matrix
beatAML_incidence_matrix <- nominalAsBinet(beatAML_data)

# plot with the interactive bipartite network

plotBipartiteInteractive(inc_mat = beatAML_incidence_matrix)

Plot the clusters in one projection of the bipartite network

Description

This function makes an interactive network figure that shows nodes in which nodes belonging to the same cluster are colored the same and nodes belonging to other clusters are colored differently. This function has been customized to use the output of findCluster.

Usage

plotCluster(graph, cluster, ...)

Arguments

Value

A networkD3 object.

See Also

forceNetwork

Examples

# generate an incidence matrix
data <- matrix(c(1, 0, 0, 0, 1, 1, 1, 0, 0, 0, 1, 1, 1, 0, 0), nrow = 3)
colnames(data) <- letters[1:5]
rownames(data) <- LETTERS[1:3]

# run findCluster() to do clustering
cls <- findCluster(
  data,
  part = 1,
  method = "all",
  normalization = FALSE,
  rm_weak_edges = TRUE,
  comparison = FALSE
)
# plot the cluster with Louvain method
plotCluster(graph = cls$graph, cluster = cls$louvain)

Plot the incidence matrix.

Description

This function converts a nominal data edge list to an incidence matrix and provides some information about that.

Usage

plotIncMatrix(
  x,
  index_nominal = c(1, 2),
  index_numeric = NULL,
  palette = "Blues",
  verbose = FALSE,
  plot_weight = FALSE,
  print_skim = FALSE
)

Arguments

Details

This function mainly converts data in the form of edge list into matrix data. It also returns the incidence matrix object, the dimensions and proportion of missing values, as well as the matrix's image for visualization.

Value

An incidence matrix object and image with the dimension and missing value proportion.

See Also

nominalAsBinet

Examples

# load part of the beatAML data
beatAML_data <- NIMAA::beatAML[1:1000,]

# visualize the input dataset
beatAML_incidence_matrix <- plotIncMatrix(beatAML_data,
index_numeric= 3)

Edge prediction of weighted bipartite network.

Description

This function utilizes several data imputation methods in order to predict the existence of a link between two nodes by imputing the edges' weight in a weighted bipartite network of nominal data.

Usage

predictEdge(inc_mat, method = c("svd", "median", "als", "CA"))

Arguments

Details

This function performs a variety of numerical imputations according to the user's input, and returns a list of imputed data matrices based on each method separately, such as median which replaces the missing values with the median of each rows (observations), and knn which uses the k-Nearest Neighbour algorithm to impute missing values.

Value

A list of matrices with original and imputed values by different methods.

See Also

knn.impute, softImpute, imputeCA, imputeFAMD, imputePCA, mice.

Examples

# load part of the beatAML data
beatAML_data <- NIMAA::beatAML[1:10000,]

# convert to incidence matrix
beatAML_incidence_matrix <- nominalAsBinet(beatAML_data)

# predict the edges by imputation the weights
predictEdge(beatAML_incidence_matrix)

robertson

Description

Charles Robertson's detailed collections provide as an amazing resource for long-term comparisons of numerous terrestrial plant and insect species collected in a single location. From 1884 through 1916, Robertson observed and collected 1429 insect species that visited 456 flower of plant species. He gathered this dataset in Macoupin County, Illinois, USA, within a 16-kilometer radius of Carlinville.

Usage

robertson

Format

A tibble with two variables:

plant

Scientific Latin names of plants

pollinator

Scientific Latin names of pollinators for the plant

Source

Marlin, J. C., & LaBerge, W. E. (2001). The native bee fauna of Carlinville, Illinois, revisited after 75 years: a case for persistence. Conservation Ecology, 5(1). http://www.ecologia.ib.usp.br/iwdb/html/robertson_1929.html

Score the cluster analysis in a projected network based on additional internal measures.

Description

This function provides additional internal cluster validity measures such as entropy and coverage. The concept of scoring is according to the weight fraction of all intra-cluster edges relative to the total weights of all edges in the graph. This function requires the community object, igraph object and distance matrix returned by findCluster to analyze.

Usage

scoreCluster(community, graph, dist_mat)

Arguments

Value

A list containing internal cluster validity scores.

See Also

cluster.stats, findCluster

Examples

# load part of the beatAML data
beatAML_data <- NIMAA::beatAML[1:10000,]

# convert to incidence matrix
beatAML_incidence_matrix <- nominalAsBinet(beatAML_data)

# do clustering
cls <- findCluster(beatAML_incidence_matrix,
  part = 1, method = "infomap", normalization = FALSE,
  rm_weak_edges = TRUE, comparison = FALSE)

# get the scoring result
scoreCluster(community = cls$infomap, graph = cls$graph,
  dist_mat = cls$distance_matrix)

Validate the cluster analysis in a projected network based on additional external measures.

Description

This function calculates the similarity of a given clustering method to the provided ground truth as external features (prior knowledge). This function provides external cluster validity measures including corrected.rand and ⁠jaccard similarity⁠. This function requires the community object, igraph object and distance matrix returned by findCluster to analyze.

Usage

validateCluster(community, extra_feature, dist_mat)

Arguments

Value

A list containing the similarity measures for the clustering results and the ground truth represented as an external features, i.e., corrected Rand and Jaccard indices.

Examples

# load part of the beatAML data
beatAML_data <- NIMAA::beatAML[1:10000,]

# convert to incidence matrix
beatAML_incidence_matrix <- nominalAsBinet(beatAML_data)

# do clustering
cls <- findCluster(beatAML_incidence_matrix,
  part = 1, method = c('infomap','walktrap'),
  normalization = FALSE, rm_weak_edges = TRUE,
  comparison = FALSE)

# generate a random external_feature
external_feature <- data.frame(row.names = cls$infomap$names)
external_feature[,'membership'] <- paste('group',
sample(c(1,2,3,4), nrow(external_feature),
replace = TRUE))

# validate clusters using random external feature
validateCluster(community = cls$walktrap,
extra_feature = external_feature,
dist_mat = cls$distance_matrix)

Validate and compare edge prediction methods.

Description

This function compares the imputation approaches for predicting edges using the clustering result of a submatrix with non-missing values as a benchmark. This function performs the same analysis as the findCluster function on every imputed incidence matrices independently. Then, using different similarity measures, all imputation approaches are compared to each other, revealing how edge prediction methods affects network communities (clusters). The best method should result in a higher degree of similarity (common node membership) to the non-missing submatrix as a benchmark.

Usage

validateEdgePrediction(imputation, refer_community, clustering_args)

Arguments

Value

A list containing the following indices: Jaccard similarity, Dice similarity coefficient, Rand index, Minkowski (inversed), and Fowlkes-Mallows index. The higher value indicates the greater similarity between the imputed dataset and the benchmark.

Examples

# load part of the beatAML data
beatAML_data <- NIMAA::beatAML[1:10000,]

# convert to incidence matrix
beatAML_incidence_matrix <- nominalAsBinet(beatAML_data)

# do clustering
cls <- findCluster(beatAML_incidence_matrix, part = 1)

# predict the edges by imputation the weights
imputed_beatAML <- predictEdge(beatAML_incidence_matrix)

# validate the edge prediction
validateEdgePrediction(imputation = imputed_beatAML,
refer_community = cls$fast_greedy,
clustering_args = cls$clustering_args
)

Plot the bipartite graph with color coding for different clusters in both parts

Description

The Sankey diagram is used to depict the connections between clusters within each part of the bipartite network. The display is also interactive, and by grouping nodes within each cluster as "summary" nodes, this function emphasizes how clusters of each part are connected together.

Usage

visualClusterInBipartite(
  data,
  community_left,
  community_right,
  name_left = "Left",
  name_right = "Right"
)

Arguments

Value

A customized Sankey plot with a data frame containing the cluster pairwise relationship with the sum of weight values in the weighted bipartite network.

See Also

plot_ly

Examples

# load part of the beatAML data
beatAML_data <- NIMAA::beatAML[1:1000,]

# convert to incidence matrix
beatAML_incidence_matrix <- nominalAsBinet(beatAML_data)

# extract the Recetengular_element_max submatrix
sub_matrices <- extractSubMatrix(beatAML_incidence_matrix,
col.vars = "patient_id", row.vars = "inhibitor",
shape = c("Rectangular_element_max"))

# do clustering analysis
cls1 <- findCluster(sub_matrices$Rectangular_element_max,
part = 1, comparison = FALSE)

cls2 <- findCluster(sub_matrices$Rectangular_element_max,
part = 2, comparison = FALSE)

visualClusterInBipartite(data = beatAML_data,
community_left = cls2$leading_eigen,
community_right = cls1$fast_greedy,
name_left = 'patient_id',
name_right = 'inhibitor')