Clustering and metaclustering
Timothy Keyes
2024-08-25
Source:vignettes/clustering.Rmd
clustering.RmdOften, clustering single-cell data to identify communities of cells with shared characteristics is a major goal of high-dimensional cytometry data analysis.
To do this, tidytof provides the
tof_cluster() verb. Several clustering methods are
implemented in tidytof, including the following:
Each of these methods are wrapped by tof_cluster().
Clustering with tof_cluster()
To demonstrate, we can apply the PhenoGraph clustering algorithm to
tidytof’s built-in phenograph_data. Note
that phenograph_data contains 3000 total cells (1000 each
from 3 clusters identified in the original PhenoGraph
publication). For demonstration purposes, we also metacluster our
PhenoGraph clusters using k-means clustering.
data(phenograph_data)
set.seed(203L)
phenograph_clusters <-
phenograph_data |>
tof_preprocess() |>
tof_cluster(
cluster_cols = starts_with("cd"),
num_neighbors = 50L,
distance_function = "cosine",
method = "phenograph"
) |>
tof_metacluster(
cluster_col = .phenograph_cluster,
metacluster_cols = starts_with("cd"),
num_metaclusters = 3L,
method = "kmeans"
)
phenograph_clusters |>
dplyr::select(sample_name, .phenograph_cluster, .kmeans_metacluster) |>
head()
#> # A tibble: 6 × 3
#> sample_name .phenograph_cluster .kmeans_metacluster
#> <chr> <chr> <chr>
#> 1 H1_PhenoGraph_cluster1 6 2
#> 2 H1_PhenoGraph_cluster1 1 2
#> 3 H1_PhenoGraph_cluster1 6 2
#> 4 H1_PhenoGraph_cluster1 6 2
#> 5 H1_PhenoGraph_cluster1 6 2
#> 6 H1_PhenoGraph_cluster1 6 2The outputs of both tof_cluster() and
tof_metacluster() are a tof_tbl identical to
the input tibble, but now with the addition of an additional column (in
this case, “.phenograph_cluster” and “.kmeans_metacluster”) that encodes
the cluster id for each cell in the input tof_tbl. Note
that all output columns added to a tibble or tof_tbl by
tidytof begin with a full-stop (“.”) to reduce the
likelihood of collisions with existing column names.
Because the output of tof_cluster is a
tof_tbl, we can use dplyr’s count
method to assess the accuracy of our clustering procedure compared to
the original clustering from the PhenoGraph paper.
phenograph_clusters |>
dplyr::count(phenograph_cluster, .kmeans_metacluster, sort = TRUE)
#> # A tibble: 4 × 3
#> phenograph_cluster .kmeans_metacluster n
#> <chr> <chr> <int>
#> 1 cluster2 3 1000
#> 2 cluster3 1 1000
#> 3 cluster1 2 995
#> 4 cluster1 1 5Here, we can see that our clustering procedure groups most cells from the same PhenoGraph cluster with one another (with a small number of mistakes).
To change which clustering algorithm tof_cluster uses,
alter the method flag.
# use the kmeans algorithm
phenograph_data |>
tof_preprocess() |>
tof_cluster(
cluster_cols = contains("cd"),
method = "kmeans"
)
# use the flowsom algorithm
phenograph_data |>
tof_preprocess() |>
tof_cluster(
cluster_cols = contains("cd"),
method = "flowsom"
)To change the columns used to compute the clusters, change the
cluster_cols flag. And finally, if you want to return a
one-column tibble that only includes the cluster labels (as
opposed to the cluster labels added as a new column to the input
tof_tbl), set augment to
FALSE.
# will result in a tibble with only 1 column (the cluster labels)
phenograph_data |>
tof_preprocess() |>
tof_cluster(
cluster_cols = contains("cd"),
method = "kmeans",
augment = FALSE
) |>
head()
#> # A tibble: 6 × 1
#> .kmeans_cluster
#> <chr>
#> 1 2
#> 2 1
#> 3 19
#> 4 9
#> 5 2
#> 6 9Session info
sessionInfo()
#> R version 4.4.1 (2024-06-14)
#> Platform: x86_64-pc-linux-gnu
#> Running under: Ubuntu 22.04.4 LTS
#>
#> Matrix products: default
#> BLAS: /usr/lib/x86_64-linux-gnu/openblas-pthread/libblas.so.3
#> LAPACK: /usr/lib/x86_64-linux-gnu/openblas-pthread/libopenblasp-r0.3.20.so; LAPACK version 3.10.0
#>
#> locale:
#> [1] LC_CTYPE=C.UTF-8 LC_NUMERIC=C LC_TIME=C.UTF-8
#> [4] LC_COLLATE=C.UTF-8 LC_MONETARY=C.UTF-8 LC_MESSAGES=C.UTF-8
#> [7] LC_PAPER=C.UTF-8 LC_NAME=C LC_ADDRESS=C
#> [10] LC_TELEPHONE=C LC_MEASUREMENT=C.UTF-8 LC_IDENTIFICATION=C
#>
#> time zone: UTC
#> tzcode source: system (glibc)
#>
#> attached base packages:
#> [1] stats graphics grDevices utils datasets methods base
#>
#> other attached packages:
#> [1] dplyr_1.1.4 tidytof_0.99.8
#>
#> loaded via a namespace (and not attached):
#> [1] tidyselect_1.2.1 viridisLite_0.4.2 timeDate_4032.109
#> [4] farver_2.1.2 viridis_0.6.5 ggraph_2.2.1
#> [7] fastmap_1.2.0 tweenr_2.0.3 rpart_4.1.23
#> [10] digest_0.6.37 timechange_0.3.0 lifecycle_1.0.4
#> [13] yardstick_1.3.1 survival_3.6-4 magrittr_2.0.3
#> [16] compiler_4.4.1 rlang_1.1.4 sass_0.4.9
#> [19] tools_4.4.1 igraph_2.0.3 utf8_1.2.4
#> [22] yaml_2.3.10 data.table_1.15.4 knitr_1.48
#> [25] graphlayouts_1.1.1 htmlwidgets_1.6.4 withr_3.0.1
#> [28] purrr_1.0.2 RProtoBufLib_2.16.0 BiocGenerics_0.50.0
#> [31] desc_1.4.3 nnet_7.3-19 grid_4.4.1
#> [34] polyclip_1.10-7 stats4_4.4.1 fansi_1.0.6
#> [37] RcppHNSW_0.6.0 future_1.34.0 colorspace_2.1-1
#> [40] ggplot2_3.5.1 globals_0.16.3 scales_1.3.0
#> [43] iterators_1.0.14 MASS_7.3-60.2 cli_3.6.3
#> [46] rmarkdown_2.28 ragg_1.3.2 generics_0.1.3
#> [49] future.apply_1.11.2 tzdb_0.4.0 cachem_1.1.0
#> [52] flowCore_2.16.0 ggforce_0.4.2 stringr_1.5.1
#> [55] splines_4.4.1 parallel_4.4.1 matrixStats_1.3.0
#> [58] vctrs_0.6.5 hardhat_1.4.0 glmnet_4.1-8
#> [61] Matrix_1.7-0 jsonlite_1.8.8 cytolib_2.16.0
#> [64] hms_1.1.3 S4Vectors_0.42.1 ggrepel_0.9.5
#> [67] listenv_0.9.1 systemfonts_1.1.0 foreach_1.5.2
#> [70] gower_1.0.1 tidyr_1.3.1 jquerylib_0.1.4
#> [73] recipes_1.1.0 parallelly_1.38.0 glue_1.7.0
#> [76] pkgdown_2.1.0 codetools_0.2-20 stringi_1.8.4
#> [79] lubridate_1.9.3 gtable_0.3.5 shape_1.4.6.1
#> [82] munsell_0.5.1 tibble_3.2.1 pillar_1.9.0
#> [85] htmltools_0.5.8.1 ipred_0.9-15 lava_1.8.0
#> [88] R6_2.5.1 textshaping_0.4.0 doParallel_1.0.17
#> [91] tidygraph_1.3.1 evaluate_0.24.0 Biobase_2.64.0
#> [94] lattice_0.22-6 readr_2.1.5 memoise_2.0.1
#> [97] bslib_0.8.0 class_7.3-22 Rcpp_1.0.13
#> [100] prodlim_2024.06.25 gridExtra_2.3 xfun_0.47
#> [103] fs_1.6.4 pkgconfig_2.0.3