General network analysis on Adaptive Immune Receptor Repertoire
Sequencing (AIRR-Seq) data is performed using
buildRepSeqNetwork()
or its convenient alias
buildNet()
. This function does the following:
igraph
and adjacency matrix)We simulate some toy data for demonstration.
We simulate data consisting of two samples with 100 observations each, for a total of 200 observations (rows).
set.seed(42)
library(NAIR)
#> Welcome to NAIR: Network Analysis of Immune Repertoire.
#> Get started using `vignette("NAIR")`, or by visiting
#> https://mlizhangx.github.io/Network-Analysis-for-Repertoire-Sequencing-/
dir_out <- tempdir()
toy_data <- simulateToyData()
head(toy_data)
#> CloneSeq CloneFrequency CloneCount SampleID
#> 1 TTGAGGAAATTCG 0.007873775 3095 Sample1
#> 2 GGAGATGAATCGG 0.007777102 3057 Sample1
#> 3 GTCGGGTAATTGG 0.009094910 3575 Sample1
#> 4 GCCGGGTAATTCG 0.010160859 3994 Sample1
#> 5 GAAAGAGAATTCG 0.009336593 3670 Sample1
#> 6 AGGTGGGAATTCG 0.010369470 4076 Sample1
data
accepts a data frame
containing the AIRR-seq data, where each row corresponds to a single
TCR/BCR clone (bulk data) or cell (single-cell data).seq_col
specifies the data column
containing the receptor sequences to be used as the basis of similarity
between two cells or clones. It accepts the column name as a character
string or the column position index.The following options are useful for removing noise or irrelevant data from the analysis, potentially improving the quality of the network graph and downstream analysis results.
The min_seq_length
parameter specifies the minimum
TCR/BCR sequence length (in number of characters). Sequences with
lengths below this value are dropped prior to analysis. An argument
value of NULL
bypasses this filter. This filter applies to
the sequence in the column specified by seq_col
.
By default, min_seq_length
is set to 3.
The optional drop_matches
parameter can be used to
remove TCR/BCR sequences matching a specified pattern. It accepts a
character string containing a regular expression specifying the pattern
of content to search for. Each sequence in the column specified by seq_col
is checked for a pattern match
using grep()
. If a match is found, the sequence is removed
prior to analysis.
The following settings pertain to the network construction.
The default method for quantifying the similarity between TCR/BCR
sequences is the Hamming distance, which is computed using
hamDistBounded()
. It calculates the number of differences
between two sequences of the same length. For each position in one
sequence, the character in that position is checked to see whether it
differs from the character in the same position of the other sequence.
If the sequences have different lengths, the shorter sequence is
extended by adding non-matching characters to make it the same length as
the longer sequence.
The Levenshtein distance, which is computed using
levDistBounded()
, can be used as an alternative measurement
to determine the similarity between sequences. It calculates the minimum
number of single-character edits (insertions, deletions and
transformations) needed to transform one sequence into the other. This
method is particularly useful for comparing sequences of different
lengths and can account for insertions and deletions. When constructing
a network based on the similarity of CDR-3 nucleotide sequences, it is
preferable to use the Levenshtein distance instead of the default
Hamming distance by specifying dist_type = "lev"
. However,
the Levenshtein distance requires significantly more computation than
the Hamming distance, which may be challenging when working with large
data sets having long TCR/BCR sequences.
The function specified by dist_type
quantifies the
similarity between TCR/BCR sequences as a nonnegative distance, with
values closer to 0 indicating greater similarity.
Each node in the network graph corresponds to a row of the AIRR-Seq
data. By default, two nodes are connected by an edge if the distance
between their TCR/BCR sequences does not exceed 1. This cutoff value is
specified by the dist_cutoff
parameter, which controls the
stringency of the network construction and affects the number and
density of edges in the network. A lower cutoff requires greater
sequence similarity to form an edge connection. If
dist_cutoff = 0
, two sequences must be identical in order
for their nodes to be joined by an edge.
buildRepSeqNetwork()
can perform additional analysis
after constructing the network, including cluster analysis (partitioning
the network graph into densely-connected subgraphs) and computation of
network properties (which describe the structural organization of the
network).
Node-level network properties pertain to individual nodes in the network graph.
Some are local properties, meaning that their value for a given node depends only on a subset of the nodes in the network. One example is the network degree of a given node, which represents the number of other nodes that are directly joined to the given node by an edge connection.
Other properties are global properties, meaning that their value for a given node depends on all of the nodes in the network. An example is the authority score of a node, which is computed using the entire graph adjacency matrix (if we denote this matrix by \(A\), then the principal eigenvector of \(A^T A\) represents the authority scores of the network nodes).
Node-level network properties can be computed when calling
buildRepSeqNetwork()
by setting
node_stats = TRUE
, or as a separate step using
addNodeStats()
.
See this vignette for more details on computing node-level network properties.
Cluster analysis uses a community-finding algorithm to partition the network graph into clusters (densely-connected subgraphs). These clusters represent groups of clones/cells with similar receptor sequences.
Cluster analysis can be performed when calling
buildRepSeqNetwork()
by setting
cluster_stats = TRUE
or as a separate step using
addClusterStats()
.
The cluster membership of each node is recorded as a variable in the node metadata. Cluster properties, such as node count and mean sequence length, are included in their own data frame within the output.
See this vignette for more details.
Customized plots can be produced when calling
buildRepSeqNetwork()
or created afterward using
addPlots()
.
Specify print_plots = TRUE
to print the plots to the R
plotting window.
net <- buildNet(toy_data, "CloneSeq",
node_stats = TRUE,
color_nodes_by = c("SampleID", "transitivity", "coreness"),
color_scheme = c("default", "plasma-1", "mako-1"),
color_title = c("", "Transitivity", "Coreness"),
size_nodes_by = "degree",
node_size_limits = c(0.1, 1.5),
plot_title = NULL,
plot_subtitle = NULL,
print_plots = TRUE
)
See this article for more on how to customize the visualizations.
The function returns a list with the following elements:
We describe each element below.
details
records argument values supplied in the call to
buildRepSeqNetwork()
:
net$details
#> $seq_col
#> [1] "CloneSeq"
#>
#> $dist_type
#> [1] "hamming"
#>
#> $dist_cutoff
#> [1] 1
#>
#> $drop_isolated_nodes
#> [1] TRUE
#>
#> $nodes_in_network
#> [1] 122
#>
#> $min_seq_length
#> [1] 3
#>
#> $drop_matches
#> [1] "NULL"
nodes_in_network
records the number of nodes in the
network. Additional network information is included when cluster analysis is performed.
node_data
is a data frame containing metadata for the
network nodes, where each row corresponds to a node in the network
graph:
head(net$node_data)
#> CloneSeq CloneFrequency CloneCount SampleID degree transitivity
#> 2 GGAGATGAATCGG 0.007777102 3057 Sample1 1 NaN
#> 5 GAAAGAGAATTCG 0.009336593 3670 Sample1 3 0.3333333
#> 8 GGGGAGAAATTGG 0.006220155 2445 Sample1 2 1.0000000
#> 11 GGGGGAGAATTGC 0.012969469 5098 Sample1 4 0.6666667
#> 12 GGGGGGGAATTGC 0.009079646 3569 Sample1 10 0.3555556
#> 13 AGGGGGAAATTGG 0.014941093 5873 Sample1 5 0.1000000
#> eigen_centrality centrality_by_eigen betweenness centrality_by_betweenness
#> 2 0.00000000 0.00000000 0.000000 0.000000
#> 5 0.00000000 0.00000000 108.000000 108.000000
#> 8 0.04558649 0.04558649 0.000000 0.000000
#> 11 0.15055366 0.15055366 1.549451 1.549451
#> 12 0.52691798 0.52691798 73.970918 73.970918
#> 13 0.14682343 0.14682343 75.439560 75.439560
#> authority_score coreness page_rank
#> 2 0.00000000 1 0.008196721
#> 5 0.00000000 2 0.010578507
#> 8 0.04558649 2 0.003936684
#> 11 0.15055366 4 0.005034736
#> 12 0.52691798 6 0.011491588
#> 13 0.14682343 3 0.008703523
Since buildNet()
was called with drop_isolated_nodes = TRUE
, some
of the original rows are missing. The original row names are preserved,
facilitating cross-referencing with the original data.
By default, all variables from the original input data are present.
To include only a subset of the original variables, specify those to
keep using the subset_cols
parameter, which accepts a
character vector of column names or a vector of column indices. The
sequence column is always included.
Variables for node-level network properties will also be present if computed, as seen here.
Variables present in the node metadata can be used to encode node colors in visualizations.
If buildRepSeqNetwork()
is called with cluster_stats = TRUE
, the
returned list will contain a data frame cluster_data
with
cluster properties such as node count and mean sequence length, as seen
here.
plots
is a list containing any plots created as well as
a matrix graph_layout
storing the node layout used in the
plots. Each plot is named according to the variable used to color the
nodes.
Each plot is a ggraph
(a
special kind of ggplot
).
Custom plots can be created when calling buildNet()
or
afterward using addPlots()
as described here.
igraph
is an object of class igraph
and
adjacency_matrix
is a dgCMatrix
.
Both objects encode the nodes and edges of the network graph, each using
a different format. These objects are used by other NAIR
functions, and are typically not of direct importance to the user.
buildRepSeqNetwork()
writes its output to files if a
directory path is provided to the output_dir
parameter. The
specified output directory will be created if it does not already
exist.
buildRepSeqNetwork()
is saved as a compressed RDS file.output_type = "rda"
. The list will be named
net
in the R environment.output_type = "individual"
. Each list element will be saved
as a separate file using the following formats:
.csv
for node_data
and
cluster_data
(using write.csv()
, with
row.names = FALSE
for cluster_data
).mtx
for adjacency_matrix
.txt
for details
, igraph
and
plots$graph_layout
.rda
(RData) for plots
output_type
, plots are printed to a pdf
file containing one plot per page. The dimensions (in inches) for each
page can be adjusted using the plot_width
and
plot_height
parameters, with the defaults being
12
and 10
, respectively.By default, the name of each saved file begins with
MyRepSeqNetwork
. This prefix can be changed using the
output_name
parameter, which accepts a character
string.
When output_type
is "rds"
or
"rda"
, only two files are saved (the data file and the
pdf). For each file, output_name
is followed by the
appropriate file extension. For example, if
output_name = "NetworkABC"
and
output_type = "rds"
, then the files will be named
NetworkABC.rds
and NetworkABC.pdf
.
output_type = "individual"
, the prefix specified by
output_name
is followed by:
_Details.txt
for details
_EdgeList.txt
for igraph
_AdjacencyMatrix.mtx
for
adjacency_matrix
_NodeMetadata.csv
for node_data
_ClusterMetadata.csv
for cluster_data
(if
present)_Plots.rda
for the list plots
.pdf
for the pdf of the plots_GraphLayout.txt
for
plots$graph_layout
For example, if output_name = "NetworkABC"
and
output_type = "individual"
, then the node metadata is saved
as NetworkABC_NodeMetadata.csv
, the pdf is saved as
NetworkABC.pdf
, the list plots
is saved as
NetworkABC_Plots.rds
, and so on.
The NAIR
package contains a number of functions
supplementary to buildRepSeqNetwork()
that can be used to
perform additional downstream tasks. See this vignette.
library(magrittr) # For pipe operator (%>%)
toy_data %>%
filterInputData("CloneSeq", drop_matches = "\\W") %>%
buildNet("CloneSeq") %>%
addNodeStats("all") %>%
addClusterMembership("greedy", cluster_id_name = "cluster_greedy") %>%
addClusterMembership("leiden", cluster_id_name = "cluster_leiden") %>%
addClusterStats("cluster_leiden", "CloneSeq", "CloneCount") %>%
addPlots(color_nodes_by = c("cluster_leiden", "cluster_greedy"),
color_scheme = "Viridis"
) %>%
labelClusters("cluster_leiden", cluster_id_col = "cluster_leiden") %>%
labelClusters("cluster_greedy", cluster_id_col = "cluster_greedy") %>%
saveNetwork(output_dir = tempdir(), output_name = "my_network")