Extended Data Figure 8: Cluster-level GP membership within each T cell lineage

workflowr

• Summary
• Checks
• Past versions

Last updated: 2026-08-27

Checks: 7 0

Knit directory: immgenT-GP-analysis/analysis/

This reproducible R Markdown analysis was created with workflowr (version 1.7.2). The Checks tab describes the reproducibility checks that were applied when the results were created. The Past versions tab lists the development history.

---

R Markdown file: up-to-date

Great! Since the R Markdown file has been committed to the Git repository, you know the exact version of the code that produced these results.

Environment: empty

Great job! The global environment was empty. Objects defined in the global environment can affect the analysis in your R Markdown file in unknown ways. For reproduciblity it’s best to always run the code in an empty environment.

Seed: set.seed(1)

The command set.seed(1) was run prior to running the code in the R Markdown file. Setting a seed ensures that any results that rely on randomness, e.g. subsampling or permutations, are reproducible.

Session information: recorded

Great job! Recording the operating system, R version, and package versions is critical for reproducibility.

Cache: none

Nice! There were no cached chunks for this analysis, so you can be confident that you successfully produced the results during this run.

File paths: relative

Great job! Using relative paths to the files within your workflowr project makes it easier to run your code on other machines.

Repository version: 659ff00

Great! You are using Git for version control. Tracking code development and connecting the code version to the results is critical for reproducibility.

The results in this page were generated with repository version 659ff00. See the Past versions tab to see a history of the changes made to the R Markdown and HTML files.

Note that you need to be careful to ensure that all relevant files for the analysis have been committed to Git prior to generating the results (you can use wflow_publish or wflow_git_commit). workflowr only checks the R Markdown file, but you know if there are other scripts or data files that it depends on. Below is the status of the Git repository when the results were generated:

Ignored files:
    Ignored:    .DS_Store
    Ignored:    .claude/
    Ignored:    analysis/.DS_Store
    Ignored:    analysis/.Rhistory
    Ignored:    analysis/assets/.DS_Store
    Ignored:    captions/
    Ignored:    code/.DS_Store
    Ignored:    code/other/topic_flashier_20250212.R
    Ignored:    code/other/topic_wrapper_20250215_alldata_backfit.sh
    Ignored:    data
    Ignored:    experiments/
    Ignored:    figures/.DS_Store
    Ignored:    figures/Previous/.DS_Store
    Ignored:    figures/Previous/bits/.DS_Store
    Ignored:    figures/Previous/bits/Figure 1/.DS_Store
    Ignored:    figures/Previous/bits/Figure 2/.DS_Store
    Ignored:    figures/Previous/bits/Figure 3/.DS_Store
    Ignored:    figures/Previous/bits/Figure 4/.DS_Store
    Ignored:    figures/Previous/bits/Figure 6/.DS_Store
    Ignored:    figures/Previous/bits/Figure 7/.DS_Store
    Ignored:    figures/Previous/bits/Figure S1/.DS_Store
    Ignored:    figures/Previous/bits/Figure S2/.DS_Store
    Ignored:    figures/Previous/bits/Figure S3/.DS_Store
    Ignored:    figures/Previous/bits/Figure S6/.DS_Store
    Ignored:    figures/Previous/bits/Figure S7/.DS_Store
    Ignored:    figures/final-selected/.DS_Store
    Ignored:    figures/final-selected/Figure 1/.DS_Store
    Ignored:    figures/final-selected/Figure 2/.DS_Store
    Ignored:    figures/final-selected/Figure 4/.DS_Store
    Ignored:    figures/final-selected/Figure S1/.DS_Store
    Ignored:    figures/final-selected/Figure S4/.DS_Store
    Ignored:    figures/templates_20260729/
    Ignored:    log/
    Ignored:    output/.DS_Store
    Ignored:    output/Figure2/
    Ignored:    output/Figure7b/7b_cell_metadata.csv.gz
    Ignored:    plan/
    Ignored:    tables/
    Ignored:    tmp/

Note that any generated files, e.g. HTML, png, CSS, etc., are not included in this status report because it is ok for generated content to have uncommitted changes.

---

These are the previous versions of the repository in which changes were made to the R Markdown (analysis/FigureS8.Rmd) and HTML (docs/FigureS8.html) files. If you’ve configured a remote Git repository (see ?wflow_git_remote), click on the hyperlinks in the table below to view the files as they were in that past version.

File	Version	Author	Date	Message
Rmd	659ff00	Ziang Zhang	2026-08-27	Extended Data Figure 8: per-lineage cluster structure plots

---

All seven panels are produced by script/FigureS8.R. They are the per-cluster, all-GP counterpart of Fig. 3B, which shows six hand-picked lineage-defining GPs grouped by lineage: here every GP that marks a sub-lineage cluster is shown, and cells are grouped by cluster within each lineage. The GPs are selected from the cluster AUCs published as Extended Data Table 6, and script/verify_structure_plot_gps.R re-derives every panel’s GP set from that published table, failing if it disagrees with the panels or with the caption below. The code is shown for reference (not re-executed on this page, since it loads the full cell-by-GP loading matrix); the images are its pre-rendered output.

Setup

library(ggplot2)
library(ggrastr)
library(cowplot)
library(fastTopics) # structure_plot()

if (!file.exists("code/R/structure_plot_panels.R")) {
  stop("Run this script from the immgenT-GP-analysis repository root.")
}
source("code/R/structure_plot_panels.R")

data_path <- "data/"
figure_path <- "figures/final-selected/Figure S8/"
record_path <- "output/FigureS8/"
dir.create(figure_path, recursive = TRUE, showWarnings = FALSE)
dir.create(record_path, recursive = TRUE, showWarnings = FALSE)

# A figure script here was once seen to exit 0 with a complete log and write
# nothing at all (see script/README.md, "A re-run can silently not write"), so
# record when this run started and assert at the end that the panels are newer.
run_started_at <- Sys.time()

Panel definitions and GP selection

The panel map, the AUC threshold and the display filters live in code/R/structure_plot_panels.R, which the verification script reads as well, so the figure and the check cannot disagree about what is shown:

# Panel letter per lineage, in panel order -- the level-1 lineage order of
# Figure 1D and Figure 3, with thymocytes and DP cells excluded as they are
# there. Both consumers iterate over the names of this map, so a lineage cannot
# be drawn, or checked, under another lineage's letter.
structure_plot_panels <- c(
  "CD8" = "s8a",
  "CD4" = "s8b",
  "Treg" = "s8c",
  "gdT" = "s8d",
  "CD8aa" = "s8e",
  "Tz" = "s8f",
  "DN" = "s8g"
)

# A GP is shown in a lineage's panel when its one-vs-rest AUC for predicting
# membership in at least one of that lineage's annotation_level2 clusters
# exceeds this threshold. Those AUCs are the values published as Extended Data
# Table 6 -- script/ExtendedDataTable6_GP_AUC_cluster.R reads the same file --
# which is what makes the two checkable against each other.
structure_plot_auc_threshold <- 0.9

# Display filters. These apply to the plotted cells only, never to the
# AUC-based GP selection above, which always uses every cluster of the lineage.
structure_plot_min_cluster_cells <- 100      # smaller clusters are not drawn
structure_plot_max_cells_per_cluster <- 2000 # larger clusters are subsampled

# ============================================================
# GP selection and the shared palette
# ============================================================
# Which lineage each cluster belongs to, from the metadata.
cluster_lineage <- cluster_lineage_map(level2_all, level1_all)

# verify_structure_plot_gps.R has only the published table to work from, so it
# maps clusters to lineages by their name prefix (CD8.A -> CD8) instead. Check
# that shortcut here, where the metadata-derived map is available, so the check
# cannot be re-deriving a different grouping than the figure drew.
auc_clusters <- rownames(auc_level2)
prefix_lineage <- sub("[.].*$", "", auc_clusters)
if (!identical(unname(cluster_lineage[auc_clusters]), prefix_lineage)) {
  disagree <- auc_clusters[unname(cluster_lineage[auc_clusters]) != prefix_lineage]
  stop(sprintf(
    "cluster name prefixes disagree with annotation_level1 for: %s",
    paste(disagree, collapse = ", ")
  ))
}

# One GP set per panel, and one color per GP over their union -- so a GP that
# marks clusters in two lineages keeps its color in both panels.
panel_gps <- gps_above_auc_by_lineage(auc_level2, cluster_lineage)
gp_colors <- structure_plot_gp_colors(unlist(panel_gps, use.names = FALSE))

message(sprintf(
  "%d GPs over %d panels (AUC > %.1f): %s",
  length(gp_colors), length(panel_gps), structure_plot_auc_threshold,
  paste(sprintf("%s %d", names(panel_gps), lengths(panel_gps)), collapse = ", ")
))

(a-g) Cluster-level GP membership by lineage

# ============================================================
# s8a-s8g: per-lineage structure plots
# ============================================================
# The loop iterates over the names of the panel map, so a lineage cannot be
# drawn under another lineage's letter.
cluster_records <- list()
gp_records <- list()

for (lineage in names(structure_plot_panels)) {
  panel <- structure_plot_panels[[lineage]]
  gps_lineage <- panel_gps[[lineage]]

  lineage_cells <- healthy_non_thymocyte[level1_all[healthy_non_thymocyte] == lineage]
  cluster_size <- table(droplevels(factor(level2_all[lineage_cells])))
  small_clusters <- names(cluster_size)[cluster_size < structure_plot_min_cluster_cells]
  lineage_cells <- lineage_cells[!level2_all[lineage_cells] %in% small_clusters]

  # Cap each cluster's width so one large cluster cannot crowd out the rest.
  set.seed(1234)
  keep <- unlist(lapply(
    split(seq_along(lineage_cells), level2_all[lineage_cells]),
    function(idx) {
      if (length(idx) > structure_plot_max_cells_per_cluster) {
        sample(idx, structure_plot_max_cells_per_cluster)
      } else {
        idx
      }
    }
  ))
  lineage_cells <- lineage_cells[keep]

  fit_lineage <- L_pm_filtered[lineage_cells, gps_lineage, drop = FALSE]
  grouping_lineage <- factor(level2_all[lineage_cells])

  # structure_plot() renames the colors it is given positionally, by the columns
  # of the matrix, so a palette in any other order would mislabel every bar.
  colors_lineage <- gp_colors[gps_lineage]
  if (!identical(names(colors_lineage), colnames(fit_lineage))) {
    stop(sprintf("panel %s: palette order does not match its GP columns.", panel))
  }

  set.seed(1234)
  p <- structure_plot(
    fit_lineage,
    topics = gps_lineage,
    gap = 40,
    n = 10000,
    colors = colors_lineage,
    grouping = grouping_lineage,
    ggplot_call = rasterized_structure_plot_call
  ) +
    labs(
      y = "membership",
      color = "",
      fill = "",
      title = sprintf(
        "%s (%d GPs, AUC > %.1f)",
        lineage,
        length(gps_lineage),
        structure_plot_auc_threshold
      )
    ) +
    guides(
      fill = guide_legend(ncol = 2),
      color = guide_legend(ncol = 2)
    ) +
    theme(
      plot.title = element_text(size = 11, face = "bold"),
      axis.text.x = element_text(size = 6, angle = 45, hjust = 1),
      axis.text.y = element_text(size = 9),
      axis.title = element_text(size = 10, face = "bold"),
      legend.position = "right",
      legend.key.size = unit(0.25, "cm"),
      legend.text = element_text(size = 5),
      legend.spacing.y = unit(0.02, "cm")
    )

  ggsave(
    filename = paste0(figure_path, panel, ".pdf"),
    plot = p,
    width = 11,
    height = 4,
    dpi = 300,
    limitsize = FALSE
  )

  # What this panel used, for the alignment check. Every cluster of the lineage
  # is listed, drawn or not: the GP selection above uses all of them.
  lineage_clusters <- sort(auc_clusters[prefix_lineage == lineage])
  drawn_size <- table(droplevels(factor(level2_all[lineage_cells])))
  cluster_records[[lineage]] <- data.frame(
    panel = panel,
    lineage = lineage,
    cluster = lineage_clusters,
    n_cells_healthy = as.integer(cluster_size[lineage_clusters]),
    n_cells_drawn = as.integer(ifelse(
      lineage_clusters %in% names(drawn_size),
      drawn_size[lineage_clusters],
      0L
    )),
    stringsAsFactors = FALSE
  )
  gp_records[[lineage]] <- data.frame(
    panel = panel,
    lineage = lineage,
    gp = gps_lineage,
    color = unname(gp_colors[gps_lineage]),
    max_auc_in_lineage = unname(apply(
      auc_level2[lineage_clusters, gps_lineage, drop = FALSE], 2,
      max, na.rm = TRUE
    )),
    stringsAsFactors = FALSE
  )
}

Extended Data Fig. 8a-g. Structure plots of single-cell GP activity (membership / GP loading) in healthy non-thymocyte cells, one panel per T cell lineage, with cells grouped by sub-lineage cluster (annotation_level2): (a) CD8, 16 GPs; (b) CD4, 22 GPs; (c) Treg, 11 GPs; (d) gdT, 44 GPs; (e) CD8aa, 9 GPs; (f) Tz, 10 GPs; (g) DN, 17 GPs – 69 distinct GPs across the seven panels. A GP is shown in a panel when its one-vs-rest AUC for predicting membership in at least one cluster of that lineage, from the GP’s loading, exceeds 0.9 (AUC > 0.9); those are the values of Extended Data Table 6, computed on the same healthy non-thymocyte cells. Selection uses all of a lineage’s clusters, including any too small to draw. Each vertical bar is one cell, partitioned among the GPs shown; bar height is that cell’s total membership over those GPs and is not normalized, so it exceeds 1 where a cell loads on several of them. One color is assigned to each of the 69 GPs and held fixed across panels, so a GP marking clusters in two lineages reads the same in both. Within a cluster, cells are ordered by a one-dimensional t-SNE of their memberships. Clusters with fewer than 100 cells are not drawn, larger clusters are subsampled to at most 2,000 cells, and at most 10,000 cells are drawn per panel – bar widths therefore reflect the cells drawn rather than cluster size. Thymocytes and DP cells are excluded, as in Fig. 3.

Record of what the panels drew

The GP sets, the palette, and the clusters each panel drew or omitted are written out alongside the panels, which is what script/verify_structure_plot_gps.R compares against Extended Data Table 6 and against the caption above:

# ============================================================
# What each panel drew, for the alignment check
# ============================================================
# script/verify_structure_plot_gps.R reads these two files, so that it can
# re-derive the panels' GP sets from the published Extended Data Table 6 without
# reloading the 1 GB loading matrix, and check the clusters drawn and omitted
# against the display filters above.
cluster_record <- do.call(rbind, cluster_records)
cluster_record$n_cells_healthy[is.na(cluster_record$n_cells_healthy)] <- 0L
write.csv(
  cluster_record,
  paste0(record_path, "s8_panel_clusters.csv"),
  row.names = FALSE
)
write.csv(
  do.call(rbind, gp_records),
  paste0(record_path, "s8_panel_gps.csv"),
  row.names = FALSE
)

Session information

sessionInfo()

R version 4.5.1 (2025-06-13)
Platform: aarch64-apple-darwin20
Running under: macOS Sequoia 15.6.1

Matrix products: default
BLAS:   /System/Library/Frameworks/Accelerate.framework/Versions/A/Frameworks/vecLib.framework/Versions/A/libBLAS.dylib 
LAPACK: /Library/Frameworks/R.framework/Versions/4.5-arm64/Resources/lib/libRlapack.dylib;  LAPACK version 3.12.1

locale:
[1] en_CA/en_CA/en_CA/C/en_CA/en_CA

time zone: America/Chicago
tzcode source: internal

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

loaded via a namespace (and not attached):
 [1] vctrs_0.7.3     cli_3.6.6       knitr_1.50      rlang_1.2.0    
 [5] xfun_0.55       stringi_1.8.9   otel_0.2.0      promises_1.5.0 
 [9] jsonlite_2.0.0  workflowr_1.7.2 glue_1.8.1      rprojroot_2.1.1
[13] git2r_0.36.2    htmltools_0.5.9 httpuv_1.6.16   sass_0.4.10    
[17] rmarkdown_2.30  evaluate_1.0.5  jquerylib_0.1.4 tibble_3.3.0   
[21] fastmap_1.2.0   yaml_2.3.12     lifecycle_1.0.5 whisker_0.4.1  
[25] stringr_1.6.0   compiler_4.5.1  fs_1.6.6        Rcpp_1.1.1-1.1 
[29] pkgconfig_2.0.3 later_1.4.4     digest_0.6.39   R6_2.6.1       
[33] pillar_1.11.1   magrittr_2.0.5  bslib_0.9.0     tools_4.5.1    
[37] cachem_1.1.0