Skip to contents

For single-cell data, cell-level network analysis can be performed based on joint similarity in alpha chain sequence and beta chain sequence.

We simulate some toy data to demonstrate the usage.

set.seed(42)
library(NAIR)

dat <- simulateToyData(chains = 2)
head(dat)
#>        AlphaSeq        BetaSeq Count UMIs SampleID
#> 1 TTGAGGAAATTCG TTGAGGAAATTCGG  3095    4  Sample1
#> 2 GGAGATGAATCGG  GGAGATGAATCGG  3057    6  Sample1
#> 3 GTCGGGTAATTGG GTCGGGTAATTGGG  3575    8  Sample1
#> 4 GCCGGGTAATTCG GCCGGGTAATTCGG  3994    7  Sample1
#> 5 GAAAGAGAATTCG GAAAGAGAATTCGG  3670    3  Sample1
#> 6 AGGTGGGAATTCG  AGGTGGGAATTCG  4076    5  Sample1

The input data is assumed to have the following format:

  • Each row corresponds to a unique cell
  • The data contains separate columns for alpha chain sequence and beta chain sequence

Dual-chain network analysis can be performed using buildRepSeqNetwork() (or generateNetworkObjects()) by supplying a length-2 vector to the seq_col parameter:

  • First entry should reference the column for alpha chain sequence
  • Second entry should reference the column for beta chain sequence
# Build network based on joint dual-chain similarity
network <- buildNet(dat, 
                    seq_col = c("AlphaSeq", "BetaSeq"),
                    count_col = "UMIs",
                    node_stats = TRUE, 
                    stats_to_include = "all",
                    cluster_stats = TRUE, 
                    color_nodes_by = "SampleID",
                    size_nodes_by = "UMIs",
                    node_size_limits = c(0.5, 3)
)

We print the network graph plot with labels added for the largest two clusters:

addClusterLabels(network$plots$SampleID, network, top_n_clusters = 2, size = 8)

The list returned buildRepSeqNetwork() the following items:

names(network)
#> [1] "details"          "igraph"           "adjacency_matrix" "adj_mat_a"       
#> [5] "adj_mat_b"        "node_data"        "cluster_data"     "plots"

Notice that the list contains three adjacency matrices: adjacency_matrix corresponds to the network based on joint similarity in both chain sequences, while adj_mat_a corresponds to the network based only on similarity in the alpha-chain sequence (and similarly for adj_mat_b).

The cluster-level data contains sequence-based cluster statistics for each of the alpha and beta chain sequences:

head(network$cluster_data)
#>   cluster_id node_count mean_A_seq_length mean_B_seq_length mean_degree
#> 1          1         15             12.13             12.87        2.60
#> 2          2         13             13.00             13.08        4.00
#> 3          3         16             13.00             13.94        5.81
#> 4          4         10             12.00             12.00        2.90
#> 5          5          3             13.00             14.00        1.67
#> 6          6          3             13.00             14.00        2.00
#>   max_degree A_seq_w_max_degree B_seq_w_max_degree agg_count max_count
#> 1          7       AAAAAAAAATTC      AAAAAAAAATTCG        42         6
#> 2         11      GGGGGGGAATTGG      GGGGGGGAATTGG        28         6
#> 3         12      GGGGGGGAATTGG     GGGGGGGAATTGGG        49         6
#> 4          6       AAAAAGAAATTG       AAAAAGAAATTG        39         7
#> 5          2      AGGGGAGAATTGG     AGGGGAGAATTGGG        10         5
#> 6          2      AAAAAAGAATTGC     AAAAAAGAATTGCG         4         2
#>   A_seq_w_max_count B_seq_w_max_count diameter_length global_transitivity
#> 1      AAAAAAAAATTC      AAAAAAAAATTC               6           0.2884615
#> 2     GGGGTGGAATTGG     GGGGTGGAATTGG               7           0.3802817
#> 3     GGGGAGAAATTGG    GGGGAGAAATTGGG               6           0.6328125
#> 4      AAAGAAAAATTG      AAAGAAAAATTG               6           0.3750000
#> 5     AGGGGAGAATTGG    AGGGGAGAATTGGG               3           0.0000000
#> 6     AGAAAAGAATTGC    AGAAAAGAATTGCG               2           1.0000000
#>   assortativity edge_density degree_centrality_index closeness_centrality_index
#> 1   -0.16503588    0.1809524               0.3190476                  0.4497821
#> 2   -0.15180055    0.2692308               0.3141026                  0.4357891
#> 3   -0.08424855    0.3416667               0.3250000                  0.4650078
#> 4   -0.33425414    0.3111111               0.3555556                  0.4889192
#> 5   -1.00000000    0.6666667               0.3333333                  1.0000000
#> 6           NaN    1.0000000               0.0000000                  0.0000000
#>   eigen_centrality_index eigen_centrality_eigenvalue
#> 1              0.6385488                    3.680389
#> 2              0.6131393                    4.419380
#> 3              0.5291669                    7.257172
#> 4              0.6107669                    3.750958
#> 5              0.5857864                    1.414214
#> 6              0.0000000                    2.000000

The remainder of the output and customization follows the general case for buildRepSeqNetwork().