Last updated: 2026-09-10

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: 8eacf9c

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 8eacf9c. 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/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 5/.DS_Store
    Ignored:    figures/final-selected/Figure S1/.DS_Store
    Ignored:    internal/
    Ignored:    log/
    Ignored:    output/.DS_Store
    Ignored:    output/Figure2/
    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/Figure4.Rmd) and HTML (docs/Figure4.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 8eacf9c Ziang Zhang 2026-09-10 Figure 4b: the column colours and order of Figure 1d and ED Fig 2d
html 88ee847 Ziang Zhang 2026-09-10 Build site: Figure 4b without the miniverse clusters
Rmd 75c6f9b Ziang Zhang 2026-09-10 Figure 4b: drop the miniverse clusters, as Figure 1d and ED Fig 2d do
html 5874416 Ziang Zhang 2026-09-10 Build site: main Figure 4 inserted, Extended Data back to 1-7
Rmd 4307b28 Ziang Zhang 2026-09-10 New main Figure 4, and fold the cluster heatmap into Extended Data Figure 2
html a7a481f Ziang Zhang 2026-09-09 Build site: the rebuilt Figure 1d and the Extended Data renumbering
Rmd c233cd8 Ziang Zhang 2026-09-09 Extended Data reorganisation: split the tissue/cluster figure, renumber 3-8
html 1e88d7e Ziang Zhang 2026-09-04 Build site: published captions and titles across all 24 pages
Rmd 0267e5b Ziang Zhang 2026-09-04 Captions from the published manuscript; trim editor notes off the page code
html 19c977f Ziang Zhang 2026-09-02 Build site: Extended Data 5-7 renumbered, Figure S5 page rebuilt
html eeca07b Ziang Zhang 2026-08-05 Keep pre-refactor provenance in panel comments off the published pages
Rmd 5651d0e Ziang Zhang 2026-08-05 Extended Data tables: reorder to six, rebuild Table 1, drop internal notes
html cbcec52 Ziang Zhang 2026-07-30 Build site: Extended Data Figure naming
Rmd 66aa029 Ziang Zhang 2026-07-30 Name the Extended Data figures as published on the site
html ac650a0 Ziang Zhang 2026-07-30 Build site: Figure S5 (ex-S6a) and Figure S6 as a-f
html 62e6e83 Ziang Zhang 2026-07-29 Build site: Figure 4 without the TF-GP network
Rmd 77f3337 Ziang Zhang 2026-07-29 Drop Figure 4’s TF-GP network; re-letter 4d/4e to 4c/4d
html ae21d37 Ziang Zhang 2026-07-28 Build site: republish after the reorder commits
html d538aa2 Ziang Zhang 2026-07-28 Build site: reordered Figures 6 / S6 / S3 and the new Figure 7b page
html 029b0ae Ziang Zhang 2026-07-28 Build site.
Rmd 0f5b5da Ziang Zhang 2026-07-28 Align all figure captions with captions_20260728_final.docx
html 3fc3789 Ziang Zhang 2026-07-27 Republish all 24 pages
html 1390a03 Ziang Zhang 2026-07-27 Republish all 24 pages
html adaef21 Ziang Zhang 2026-07-27 Build site: panel fixes and PDF-derived assets
html 5b19858 Ziang Zhang 2026-07-27 Build site.
Rmd 91ee059 Ziang Zhang 2026-07-26 Select each panel’s code block by name, not by line number
html 91ee059 Ziang Zhang 2026-07-26 Select each panel’s code block by name, not by line number
Rmd 620afca Ziang Zhang 2026-07-24 Align Fig3/4/5 section titles with current numbers; reletter Fig4 to a-e
html 620afca Ziang Zhang 2026-07-24 Align Fig3/4/5 section titles with current numbers; reletter Fig4 to a-e
Rmd 2873ad2 Ziang Zhang 2026-07-16 Unify Fig 3/4/5 panel filenames with new figure numbers
html 2873ad2 Ziang Zhang 2026-07-16 Unify Fig 3/4/5 panel filenames with new figure numbers
Rmd b197499 Ziang Zhang 2026-07-16 Restructure main figures (renumber + split Figure 1)
html b197499 Ziang Zhang 2026-07-16 Restructure main figures (renumber + split Figure 1)
html 6d1c6c0 Ziang Zhang 2026-07-04 Fix TableS1 to use non-thymocyte (not healthy-restricted) AUC, matching published Table S1
html 92021bf Ziang Zhang 2026-07-02 Build site.
html 827c89b Ziang Zhang 2026-07-02 Build site.
Rmd 8ac7f9f Ziang Zhang 2026-07-02 Add data provenance notes to each script; remove conversational
Rmd f9db962 Ziang Zhang 2026-07-02 Simplify layout: drop old code/script folders, rename
html c6e5086 Ziang Zhang 2026-07-02 Build site.
html 5a79883 Ziang Zhang 2026-07-02 Build site.
Rmd 2b0e445 Ziang Zhang 2026-07-02 Fix GitHub source links to point at the new
html cf1d0ac Ziang Zhang 2026-07-02 Build site.
Rmd 06b2461 Ziang Zhang 2026-07-02 Initial commit: immgenT-GP-analysis
html 06b2461 Ziang Zhang 2026-07-02 Initial commit: immgenT-GP-analysis

All panels are produced by script/Figure4.R, which shares its GP selection and cells with Extended Data Figure 4 via code/R/structure_plot_panels.R, and its heatmap rendering with Extended Data Figure 2d via code/R/centered_mean_heatmap.R. The code below is shown for reference (not re-executed on this page); the images are its pre-rendered output.

Setup

Data loading, shared across all panels below.

suppressPackageStartupMessages({
  library(ggplot2)
  library(ggrepel)
  library(ggrastr)
  library(cowplot)
  library(dplyr)
  library(fastTopics)      # structure_plot()
  library(ComplexHeatmap)
  library(circlize)
  library(grid)
  library(ZemmourLib)      # immgent_colors
})

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")   # Extended Data Figure 4's rows, GP rule, palette
source("code/R/centered_mean_heatmap.R")   # Extended Data Figure 2d's heatmap rendering
source("code/R/level2_group_palette.R")    # EXCLUDE_LEVEL2_GROUPS, as Figure 1d and ED 2d use

data_path   <- "data/"
figure_path <- "figures/final-selected/Figure 4/"
record_path <- "output/Figure4/"
dir.create(figure_path, recursive = TRUE, showWarnings = FALSE)
dir.create(record_path, recursive = TRUE, showWarnings = FALSE)

# Figure scripts here have been seen to exit 0 having written nothing, so every
# PDF is checked against this timestamp at the end rather than against the exit
# code (script/README.md, "A re-run can silently not write").
run_started_at <- Sys.time()

# ============================================================
# Load data
# ============================================================
seurat_meta <- readRDS(paste0(
  data_path, "igt1_96_withtotalvi20260206_clean_ADTonly.Rds"
))@meta.data
L_pm_filtered <- readRDS(paste0(data_path, "L_pm_filtered.rds"))
seurat_meta_filtered <- seurat_meta[rownames(L_pm_filtered), ]
colnames(L_pm_filtered) <- gsub("^K", "GP", colnames(L_pm_filtered))
if (!all(grepl("^GP[0-9]+$", colnames(L_pm_filtered)))) {
  stop("The loading matrix does not carry K##/GP## column names.")
}

level_1_AUC_list <- readRDS(paste0(
  data_path, "level_1_AUC_list_figure_no_thymocytes_healthy.rds"
))
level_2_AUC_list <- readRDS(paste0(
  data_path, "level_2_AUC_list_figure_no_thymocytes_healthy.rds"
))
colnames(level_1_AUC_list$auc) <- gsub("^K", "GP", colnames(level_1_AUC_list$auc))
colnames(level_2_AUC_list$auc) <- gsub("^K", "GP", colnames(level_2_AUC_list$auc))

# Both AUC matrices index GPs by names taken from the loading matrix, so they
# have to be talking about the same GPs in the same order.
if (!identical(colnames(level_1_AUC_list$auc), colnames(L_pm_filtered)) ||
      !identical(colnames(level_2_AUC_list$auc), colnames(L_pm_filtered))) {
  stop("The AUC matrices and L_pm_filtered do not carry the same GP columns in the same order.")
}

level1_all <- seurat_meta_filtered$annotation_level1
level2_all <- seurat_meta_filtered$annotation_level2
healthy_non_thymocyte <- which(
  seurat_meta_filtered$condition_broad == "healthy" & level1_all != "thymocyte"
)
meta_reference <- seurat_meta_filtered[healthy_non_thymocyte, , drop = FALSE]
L_healthy <- L_pm_filtered[healthy_non_thymocyte, , drop = FALSE]
overall_mean <- colMeans(L_healthy, na.rm = TRUE)

message(sprintf(
  "%d healthy non-thymocyte cells, %d GPs, %d level-2 clusters",
  nrow(L_healthy), ncol(L_healthy), dplyr::n_distinct(meta_reference$annotation_level2)
))

(a) Lineage vs. cluster prediction

# ============================================================
# 4a: Max AUC level-1 vs max AUC level-2
# ============================================================
# Figure 6a and 6d's construction, with their category-count filters kept as
# they are there: level-1 lineages need 1000 cells, level-2 clusters 100.
level_1_small <- table(meta_reference$annotation_level1)
level_1_small <- names(level_1_small[level_1_small < 1000])
level_1_AUC <- level_1_AUC_list$auc[
  !rownames(level_1_AUC_list$auc) %in% level_1_small, , drop = FALSE
]

level_2_small <- table(meta_reference$annotation_level2)
level_2_small <- names(level_2_small[level_2_small < 100])
level_2_AUC <- level_2_AUC_list$auc[
  !rownames(level_2_AUC_list$auc) %in% level_2_small, , drop = FALSE
]

# A high AUC only means "GP predicts this category" when the category's mean
# loading is above the overall mean; the other direction is a GP the category
# lacks. Same masking as Figure 6.
category_mean <- function(labels, categories) {
  t(vapply(categories, function(cat) {
    colMeans(L_healthy[labels == cat, , drop = FALSE], na.rm = TRUE)
  }, numeric(ncol(L_healthy))))
}
level_1_positive <- sweep(
  category_mean(meta_reference$annotation_level1, rownames(level_1_AUC)),
  2, overall_mean, "-"
) > 0
level_2_positive <- sweep(
  category_mean(meta_reference$annotation_level2, rownames(level_2_AUC)),
  2, overall_mean, "-"
) > 0

masked_max <- function(auc, positive) {
  auc[!positive] <- NA
  list(
    value = apply(auc, 2, max, na.rm = TRUE),
    name = apply(auc, 2, function(x) rownames(auc)[which.max(x)])
  )
}
l1_max <- masked_max(level_1_AUC, level_1_positive)
l2_max <- masked_max(level_2_AUC, level_2_positive)

df_a <- data.frame(
  Factor = colnames(level_1_AUC),
  annotation_Level1 = l1_max$name,
  annotation_Level2 = l2_max$name[colnames(level_1_AUC)],
  Max_AUC_Level1 = l1_max$value,
  Max_AUC_Level2 = l2_max$value[colnames(level_1_AUC)],
  stringsAsFactors = FALSE
) |>
  dplyr::mutate(residual = Max_AUC_Level2 - Max_AUC_Level1)

# Figure 6a's highlight rule verbatim: AUC > 0.9 on either axis, coloured by
# which of the two maxima is the larger. Only the labelling differs -- GP names,
# not each point's top categories.
highlighted <- df_a |>
  dplyr::filter(is.finite(residual), Max_AUC_Level1 > 0.9 | Max_AUC_Level2 > 0.9) |>
  dplyr::pull(Factor)

label_above <- df_a |>
  dplyr::filter(Factor %in% highlighted, residual > 0) |>
  dplyr::mutate(
    label_text = Factor
  )
label_below <- df_a |>
  dplyr::filter(Factor %in% highlighted, residual <= 0) |>
  dplyr::mutate(
    label_text = Factor
  )
message(sprintf(
  "a: %d of %d GPs highlighted (%d above the diagonal, %d on or below)",
  length(highlighted), nrow(df_a), nrow(label_above), nrow(label_below)
))

axis_limits <- function(x, pad = 0.04) {
  x <- x[is.finite(x)]
  c(min(x) - pad, max(x) + pad)
}

p_4a <- ggplot(df_a, aes(Max_AUC_Level1, Max_AUC_Level2)) +
  geom_point(alpha = 0.3, size = 1.8) +
  geom_point(data = label_above, color = "#1f78b4", alpha = 0.8, size = 1.8) +
  geom_point(data = label_below, color = "#e31a1c", alpha = 0.8, size = 1.8) +
  geom_abline(slope = 1, intercept = 0, linetype = "dashed", color = "black") +
  coord_cartesian(
    xlim = axis_limits(df_a$Max_AUC_Level1),
    ylim = axis_limits(df_a$Max_AUC_Level2),
    expand = FALSE, clip = "off"
  ) +
  labs(
    x = "Max AUC (Level-1)", y = "Max AUC (Level-2)",
    title = "Max AUC: Level-1 vs Level-2"
  ) +
  theme_minimal(base_size = 13) +
  theme(plot.margin = margin(10, 40, 10, 40)) +
  geom_text_repel(
    seed = 42, data = label_above, aes(label = label_text), color = "#1f78b4",
    size = 2.5, lineheight = 0.85, direction = "y", nudge_x = -0.035,
    segment.color = "#1f78b4",
    arrow = arrow(length = unit(0.008, "npc"), type = "closed", angle = 20),
    force = 3, force_pull = 0.1, box.padding = 0.4, point.padding = 0.15,
    max.time = 10, max.iter = 2e4, max.overlaps = 20,
    min.segment.length = 0.01, segment.alpha = 0.7
  ) +
  geom_text_repel(
    seed = 42, data = label_below, aes(label = label_text), color = "#e31a1c",
    size = 2.5, lineheight = 0.85, direction = "y", nudge_x = 0.035,
    segment.color = "#e31a1c",
    arrow = arrow(length = unit(0.008, "npc"), type = "closed", angle = 20),
    force = 3, force_pull = 0.1, box.padding = 0.4, point.padding = 0.15,
    max.time = 10, max.iter = 2e4, max.overlaps = 20,
    min.segment.length = 0.01, segment.alpha = 0.7
  )
ggsave(paste0(figure_path, "4a.pdf"), p_4a, width = 8, height = 8, dpi = 300)

write.csv(
  df_a[order(-df_a$Max_AUC_Level2), c(
    "Factor", "Max_AUC_Level1", "annotation_Level1",
    "Max_AUC_Level2", "annotation_Level2", "residual"
  )],
  file.path(record_path, "4a_max_auc_level1_level2.csv"),
  row.names = FALSE
)

Version Author Date
5874416 Ziang Zhang 2026-09-10
7102598 Ziang Zhang 2026-07-27
6c1a613 Ziang Zhang 2026-07-27
bf7afcf Ziang Zhang 2026-07-27
ea3ecc2 Ziang Zhang 2026-07-27
620afca Ziang Zhang 2026-07-24
b197499 Ziang Zhang 2026-07-16
78e3bba Ziang Zhang 2026-07-03
06b2461 Ziang Zhang 2026-07-02

Fig. 4a. Scatterplot showing each GP’s prediction (AUC) of level 1 lineage (x-axis) versus level 2 cluster (y-axis), in healthy non-thymocyte cells. Each point is one GP; the dashed line is the identity. GPs reaching AUC > 0.9 on either axis are colored and labelled: blue where the cluster AUC is the larger of the two, red where the lineage AUC is. Most GPs sit above the identity line, predicting a cluster better than any whole lineage.

(b) The cluster-associated GPs across clusters

# ============================================================
# 4b: the same GPs as a heatmap, level-2 columns grouped by level 1
# ============================================================
cells_drawn <- sort(unlist(lineage_cells_drawn, use.names = FALSE))
labels_drawn <- droplevels(factor(level2_all[cells_drawn]))
heat_means <- mean_loading_by_group(
  L_pm_filtered[cells_drawn, gp_union, drop = FALSE], labels_drawn
)
heat_raw <- heat_means$matrix          # GPs x clusters
heat_centered <- center_by_gp_mean(heat_raw)

# Figure 1's level-1 order, minus DP and thymocytes, which these rows exclude.
level1_order <- c("CD8", "CD4", "Treg", "gdT", "CD8aa", "Tz", "DN")
level2_group_level1 <- level2_to_level1_map(
  meta_reference, colnames(heat_raw), level1_order
)
level2_group_group <- level2_to_group_map(meta_reference, colnames(heat_raw))

# Columns follow Figure 1d and Extended Data Figure 2d exactly: lineage, then
# annotation_level2_group as a contiguous block, then cluster alphabetically --
# so each lineage's ".P" cluster sits at the end of its lineage rather than
# mid-alphabet, and the level2_group bar reads as blocks. GP rows then follow
# the columns, in dominant-cluster blocks.
heat_order <- dominant_group_order(
  heat_raw,
  level2_group_block_order(
    colnames(heat_raw), level2_group_level1, level2_group_group, level1_order
  )
)

level1_palette <- ZemmourLib::immgent_colors$level1[level1_order]

centered_color_limit <- 0.2
render_centered_heatmap(
  heat_centered,
  NULL,
  "cluster (annotation_level2)",
  paste0(figure_path, "4b.pdf"),
  heat_order$row_order,
  heat_order$column_order,
  centered_color_limit,
  sprintf(
    paste0(
      "%d GPs with AUC > %.1f in some cluster; miniverse (.wM) clusters excluded\n",
      "level2 columns: level1 order (%s); level2_group blocks, alphabetical ",
      "within block; GP rows: dominant-cluster blocks"
    ),
    nrow(heat_centered), structure_plot_auc_threshold,
    paste(level1_order, collapse = ", ")
  ),
  group_level1 = level2_group_level1,
  level1_palette = level1_palette,
  group_annotation = level2_group_group,
  group_annotation_palette = LEVEL2_GROUP_COLORS[LEVEL2_GROUP_ORDER]
)

write.csv(
  data.frame(gp = rownames(heat_centered), heat_centered, check.names = FALSE),
  file.path(record_path, "4b_row_centered_mean_loading.csv"), row.names = FALSE
)
write.csv(
  data.frame(
    cluster = heat_means$counts$group,
    level1 = unname(level2_group_level1[heat_means$counts$group]),
    n_cells = heat_means$counts$n_cells
  ),
  file.path(record_path, "4b_column_cells.csv"), row.names = FALSE
)

Version Author Date
8eacf9c Ziang Zhang 2026-09-10
75c6f9b Ziang Zhang 2026-09-10
5874416 Ziang Zhang 2026-09-10
7102598 Ziang Zhang 2026-07-27
6c1a613 Ziang Zhang 2026-07-27
bf7afcf Ziang Zhang 2026-07-27
ea3ecc2 Ziang Zhang 2026-07-27
620afca Ziang Zhang 2026-07-24
b197499 Ziang Zhang 2026-07-16
78e3bba Ziang Zhang 2026-07-03
06b2461 Ziang Zhang 2026-07-02

Fig. 4b. Heatmap of mean GP activity per level 2 cluster, for the 69 GPs that reach AUC > 0.9 in at least one cluster. Each row is centered on that GP’s mean across the clusters shown (blue, below; red, above; saturating at 0.2). Columns are the 80 clusters with at least 100 healthy non-thymocyte cells, ordered by lineage (Cell Type bar), then by activation state (Level2 Group bar) as a contiguous block, then alphabetically; rows are grouped by the cluster in which the GP is most active. Colors and column order follow Figure 1d and Extended Data Fig. 2d, and as there the miniverse (.wM) clusters are excluded.

(c) Mean GP activity of the Treg clusters

# ============================================================
# 4c: one stacked bar of mean loadings per level-2 cluster, the Treg row
# ============================================================
# The Treg row of Extended Data Figure 4 with one bar per cluster, holding that
# cluster's mean loading per GP instead of one bar per cell -- the average of
# that row's bars. Same GPs (the 11 with AUC > 0.9 in some Treg cluster), same
# per-row palette, same cells, minus Treg.wM.
#
# Bars are raw mean loadings, not rescaled to a common height: that is what
# makes them the average of those bars, since structure_plot() does not
# renormalize a subset of topics either. A tall bar means the cluster carries
# more total activity over the row's GPs, not just a different mix.
#
#
# Unlike 4d and Extended Data Figure 4 this panel is not drawn at the structure
# plot's 16 x 3 in: a chart of seven bars does not need a structure plot's
# width, and Ziang asked for tall and narrow.
bar_figure_width <- 5.5  # inches
bar_figure_height <- 7

# Mean loading per cluster over the row's GPs, plus the record of every segment.
bar_means <- function(lineage) {
  gps_lineage <- panel_gps[[lineage]]
  means <- mean_loading_by_cluster(lineage_cells_drawn[[lineage]], gps_lineage)
  bar_matrix <- means$matrix
  totals <- rowSums(bar_matrix)
  if (any(!is.finite(totals)) || any(totals <= 0)) {
    stop(sprintf("row %s: a cluster has no loading at all.", lineage))
  }

  # This row's colours, assigned from the top of the palette without reference
  # to any other row -- the same per-row rule Extended Data Figure 4 uses, so
  # colour means different GPs in different rows.
  colors_lineage <- structure_plot_row_colors(gps_lineage)
  if (!identical(names(colors_lineage), colnames(bar_matrix))) {
    stop(sprintf("row %s: palette order does not match its GP columns.", lineage))
  }

  long <- data.frame(
    cluster = factor(
      rep(rownames(bar_matrix), times = ncol(bar_matrix)),
      levels = rownames(bar_matrix)
    ),
    gp = factor(
      rep(colnames(bar_matrix), each = nrow(bar_matrix)),
      levels = colnames(bar_matrix)
    ),
    value = as.vector(bar_matrix),
    stringsAsFactors = FALSE
  )

  record <- data.frame(
    panel = structure_plot_panels[[lineage]],
    lineage = lineage,
    drawn_in_4c = lineage == bar_lineage,
    cluster = as.character(long$cluster),
    n_cells = means$counts[match(long$cluster, rownames(bar_matrix))],
    gp = as.character(long$gp),
    color = unname(colors_lineage[as.character(long$gp)]),
    mean_loading = long$value,
    # The same segment as a share of its cluster's total -- the normalized view's
    # number, kept so that variant can be redrawn from the record alone.
    proportion = long$value / totals[as.character(long$cluster)],
    stringsAsFactors = FALSE
  )

  list(long = long, colors = colors_lineage, record = record, gps = gps_lineage)
}

bar_lineage <- "Treg"
bar_data <- lapply(names(structure_plot_panels), bar_means)
names(bar_data) <- names(structure_plot_panels)

drawn <- bar_data[[bar_lineage]]
p_4c <- ggplot(drawn$long, aes(cluster, value, fill = gp)) +
  geom_col(width = 0.85) +
  scale_fill_manual(values = drawn$colors) +
  scale_y_continuous(expand = expansion(mult = c(0, 0.02))) +
  labs(
    x = "", y = "mean loading", fill = "",
    title = sprintf(
      "%s (%d GPs, AUC > %.1f)", bar_lineage, length(drawn$gps),
      structure_plot_auc_threshold
    )
  ) +
  guides(fill = guide_legend(ncol = 1)) +
  cowplot::theme_cowplot(9) +
  theme(
    plot.title = element_text(size = 11, face = "bold"),
    axis.text.x = element_text(size = 7, angle = 45, hjust = 1),
    axis.text.y = element_text(size = 9),
    axis.title = element_text(size = 10, face = "bold"),
    axis.line = element_blank(),
    axis.ticks = element_blank(),
    legend.position = "right",
    legend.key.size = unit(0.3, "cm"),
    legend.text = element_text(size = 7)
  )
ggsave(
  paste0(figure_path, "4c.pdf"), p_4c,
  width = bar_figure_width, height = bar_figure_height, dpi = 300
)

# Every bar segment of all seven rows, raw and as a share of its cluster's
# total; drawn_in_4c marks the row this panel shows.
bar_record <- do.call(rbind, lapply(bar_data, `[[`, "record"))
write.csv(bar_record, file.path(record_path, "4c_mean_loading_by_cluster.csv"), row.names = FALSE)

Version Author Date
5874416 Ziang Zhang 2026-09-10
62e6e83 Ziang Zhang 2026-07-29
7102598 Ziang Zhang 2026-07-27
6c1a613 Ziang Zhang 2026-07-27
bf7afcf Ziang Zhang 2026-07-27
ea3ecc2 Ziang Zhang 2026-07-27
620afca Ziang Zhang 2026-07-24
2873ad2 Ziang Zhang 2026-07-16
b197499 Ziang Zhang 2026-07-16
78e3bba Ziang Zhang 2026-07-03
06b2461 Ziang Zhang 2026-07-02

Fig. 4c. Mean GP activity of each Treg cluster, for the 11 GPs reaching AUC > 0.9 in at least one Treg cluster. Each bar is one cluster and is stacked by GP, so its height is that cluster’s total mean loading over these GPs. Clusters with fewer than 100 healthy non-thymocyte cells, and the miniverse (.wM) cluster, are not shown.

(d) The same GPs and cells, cell by cell

# ============================================================
# 4d: Extended Data Figure 4's Treg row on its own
# ============================================================
# FigureS4.R's loop body for one lineage, unchanged including its seeds and its
# 2000-cell cap, so this is that row and not a redrawing of it.
d_lineage <- "Treg"
gps_d <- panel_gps[[d_lineage]]
cells_d <- healthy_non_thymocyte[level1_all[healthy_non_thymocyte] == d_lineage]
cluster_size_d <- table(droplevels(factor(level2_all[cells_d])))
small_d <- names(cluster_size_d)[cluster_size_d < structure_plot_min_cluster_cells]
cells_d <- cells_d[!level2_all[cells_d] %in% small_d]
cells_d <- drop_miniverse(cells_d)   # as in 4b and 4c

set.seed(1234)
keep_d <- unlist(lapply(
  split(seq_along(cells_d), level2_all[cells_d]),
  function(idx) {
    if (length(idx) > structure_plot_max_cells_per_cluster) {
      sample(idx, structure_plot_max_cells_per_cluster)
    } else {
      idx
    }
  }
))
cells_d <- cells_d[keep_d]

fit_d <- L_pm_filtered[cells_d, gps_d, drop = FALSE]
colors_d <- structure_plot_row_colors(gps_d)
if (!identical(names(colors_d), colnames(fit_d))) {
  stop("4d: palette order does not match its GP columns.")
}

set.seed(1234)
p_4d <- structure_plot(
  fit_d, topics = gps_d, gap = 40, n = 10000, colors = colors_d,
  grouping = factor(level2_all[cells_d]),
  ggplot_call = rasterized_structure_plot_call
) +
  labs(
    y = "membership", color = "", fill = "",
    title = sprintf(
      "%s (%d GPs, AUC > %.1f)", d_lineage, length(gps_d), 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(
  paste0(figure_path, "4d.pdf"), p_4d,
  width = structure_plot_width, height = structure_plot_row_height,
  dpi = 300, limitsize = FALSE
)

Version Author Date
5874416 Ziang Zhang 2026-09-10
62e6e83 Ziang Zhang 2026-07-29
7102598 Ziang Zhang 2026-07-27
6c1a613 Ziang Zhang 2026-07-27
bf7afcf Ziang Zhang 2026-07-27
ea3ecc2 Ziang Zhang 2026-07-27
620afca Ziang Zhang 2026-07-24
2873ad2 Ziang Zhang 2026-07-16
b197499 Ziang Zhang 2026-07-16
78e3bba Ziang Zhang 2026-07-03
06b2461 Ziang Zhang 2026-07-02

Fig. 4d. Structure plot of the same GPs in the same cells at single-cell resolution: each bar is one cell, colored by its GP membership and grouped by cluster. Colors match (c). Panel (c) is the per-cluster average of these bars.


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