An integrated single cell and spatial omics atlas of human prenatal development

Code ↔ Paper

19 matches between paragraphs of the paper and lines of its authors' code, computed by the harvester (lexical-v1). Click a colored paragraph or line to see its counterpart.

The 19 matches

[1] § Methods › Whole embryo: scRNAseq integration, batch correction and visualisation ↔ reproducibility/modules/module_generate/submodule_synthetic_simple/example_operations.ipynb, lines 292–373 · score 0.84 · n_neighbors, sc.pp.neighbors, sc.tl.umap, Scanpy, dimensionality, heatmaps
[2] § Methods › Community data: Realignment and gene space harmonisation ↔ idtrack/_harmonize_features.py, lines 22–82 · score 0.81 · failed conversions, Gene symbols, gene identifiers, IDTrack, conflicts, audit
[3] § Methods › Whole embryo: scRNAseq integration, batch correction and visualisation ↔ reproducibility/local/experiments/sctram_figures_only.ipynb, lines 1793–1810 · score 0.81 · n_neighbors, sc.pp.neighbors, sc.tl.umap, Scanpy, scVI, atlas
[4] § Methods › Community data: Benchmarking integration and trajectories with scIB and scTRAM › scTRAM: Trajectory-specific fidelity evaluation ↔ reproducibility/server_sync/experiments/_deprecated_suo_incremental/incremental_analysis.ipynb, lines 70–155 · score 0.81 · graph edit distance, spectral distance, trustworthiness, Spearman, concordance, stress
[5] § Methods › Community data: Benchmarking integration and trajectories with scIB and scTRAM › Integrated benchmarking framework ↔ reproducibility/server_sync/experiments/_deprecated_suo_incremental/helper/suo_scib_iterative.py, lines 25–133 · score 0.79 · min max scaled, scIB, aggregate score, batch correction, conservation, trajectory
[6] § Methods › Community data: Benchmarking integration and trajectories with scIB and scTRAM ↔ reproducibility/local/experiments/sctram_figures_only.ipynb, lines 198–252 · score 0.78 · TarDis, scPoli, scTRAM, scVI, Harmony, PCA
[7] § Methods › Community data: Integration with scVI ↔ reproducibility/local/experiments/sctram_figures_only.ipynb, lines 198–252 · score 0.76 · TarDis, scPoli, scTRAM, scVI, Harmony, PCA
[8] § Methods › Community data: Benchmarking integration and trajectories with scIB and scTRAM › Integrated benchmarking framework ↔ reproducibility/local/experiments/sctram_figures_only.ipynb, lines 254–290 · score 0.75 · scTRAM, scIB, aggregate score, batch correction, topology, conservation
[9] § Methods › Community data: Benchmarking integration and trajectories with scIB and scTRAM › scIB: Conventional integration quality assessment ↔ reproducibility/server_sync/experiments/_deprecated_suo_incremental/incremental_analysis.ipynb, lines 241–324 · score 0.70 · graph connectivity, scIB, cLISI, iLISI, isolated, silhouette
[10] § Methods › Community data: Benchmarking integration and trajectories with scIB and scTRAM › scTRAM: Trajectory-specific fidelity evaluation ↔ reproducibility/modules/module_evaluate/example_operations_adjacency.ipynb, lines 92–119 · score 0.68 · graph edit distance, spectral distance, adjacency, correlations, scores, metric
[11] § Methods › HDCA: Integration and annotation harmonisation of the final integrated community and whole embryo data ↔ reproducibility/server_sync/experiments/_deprecated_suo_incremental/helper/model_training.py, lines 15–97 · score 0.66 · gene likelihood, batch_key, trained, covariate, latent, layer
[12] § Results › Benchmarking atlas integration with trajectory fidelity ↔ reproducibility/local/experiments/sctram_figures_only.ipynb, lines 380–453 · score 0.66 · TarDis, scPoli, scVI, Harmony, PCA, scTRAM
[13] § Methods › scRNAseq gene expression and UMAP visualisation ↔ reproducibility/modules/module_generate/submodule_synthetic_simple/example_operations.ipynb, lines 292–373 · score 0.64 · sc.pl.umap, Scanpy, pp, heatmaps, matrices
[14] § Methods › HDCA: Integration and annotation harmonisation of the final integrated community and whole embryo data ↔ reproducibility/server_sync/experiments/_deprecated_suo_incremental/incremental_analysis.ipynb, lines 70–155 · score 0.63 · Normalised Mutual Information, macrophage, myeloid, concordance, Suo, neighbour
[15] § Results › Benchmarking atlas integration with trajectory fidelity ↔ reproducibility/local/experiments/sctram_figures_only.ipynb, lines 1793–1810 · score 0.58 · ground truth, UMAP, HSC, MPP, scPoli, scTRAM
[16] § Methods › Community data: Highly variable genes selection pipeline ↔ idtrack/_harmonize_features.py, lines 22–82 · score 0.53 · harmonised genes, IDTrack, single cell, logs, conversion, downstream
[17] § Methods › Community data: Public human developmental single cell data acquisition and preprocessing ↔ idtrack/_harmonize_features.py, lines 876–941 · score 0.52 · trivial, age, provenance, donor, concatenated, schema
[18] § Methods › Community data: Integration with scVI ↔ idtrack/_harmonize_features.py, lines 943–979 · score 0.51 · AnnData, Scanpy, uns, post, validation, genes
[19] § Methods › Whole embryo: Xenium preprocessing and analysis ↔ reproducibility/server_sync/experiments/_deprecated_suo_incremental/helper/model_training.py, lines 15–97 · score 0.50 · categorical covariate, unlabelled, SCANVI, latent, layers, Batch

Paper

Loaded from Europe PMC by your browser, not stored by OSCR: doi.org · Europe PMC

The paper is loaded when this pane is shown.

The paper is loaded by your browser, with JavaScript: read it at doi.org or on Europe PMC.

The authors' code

Jupyter notebook · 1,813 lines · 57 KB · BSD-3-Clause · 6 matches

# %% [markdown]
# # Imports
# %%
0
# %%
import sys
import os
import gc
import networkx as nx
import numpy as np
import pandas as pd
import tqdm
# %%
import anndata as ad
import scanpy as sc
# import scvelo as scv
# import cellrank as cr
# import scanpy.external as sce
# scv.settings.verbosity = 3
# cr.settings.verbosity = 2
# sc.settings.verbose = 3
# %%
working_directory = "/Users/kemalinecik/git_nosync/sctram"
sys.path.append(working_directory)
sys.path.append("/Users/kemalinecik/git_nosync/sctram/reproducibility/server_sync/experiments")
from incremental_training.helper.constants import metrics_scib, metric_direction_dict
from development_atlas_figure.helper.get_hdca_trainings_with_paths import get_hdca_trainings_with_paths
from development_atlas_figure.helper.pickle_operations import save_pickle_bundle, load_pickle_bundle
from development_atlas_figure.helper.csv_operations import (
harmonize_union_with_report,
analyze_column_consistency,
combine_dataframes_on_columns
)
from sctram.api._lower_level import TrajectoryEvaluationAPI
from sctram.generate.real import sc_suo_developmental_complete
from sctram.input import InputTrajectories
# %%
%matplotlib inline
%config InlineBackend.figure_format='retina'
import pickle
import matplotlib.pyplot as plt
import seaborn as sns
import colorcet as cc
import matplotlib.patheffects as path_effects
from adjustText import adjust_text
from matplotlib import gridspec
from networkx.drawing.nx_agraph import graphviz_layout
import matplotlib as mpl
import matplotlib.patches as mpatches
from matplotlib.ticker import FormatStrFormatter
from adjustText import adjust_text
import matplotlib.patheffects as path_effects
import matplotlib.cm as cm
from plottable import ColumnDefinition, Table
from plottable.cmap import normed_cmap
from matplotlib.ticker import MaxNLocator
from plottable.plots import bar
_rcparams_path = os.path.join(working_directory, "reproducibility/figure_rcparams/rcparams.pickle")
with open(_rcparams_path, "rb") as file:
_rcparams = pickle.load(file)
plt.rcParams.update(_rcparams)
plt.rcParams["font.family"] = "Arial"
plt.rcParams["font.sans-serif"] = ["Arial"]
# %%
temp_move = "/Users/kemalinecik/Downloads/temp_move"
# %% [markdown]
# how many
# %%
(
df_hdca_component_adata_paths,
df_hdca_lvl3_adata_paths,
out_df,
ground_truth_trajectories,
ground_truth_trajectories_annotation_level,
trajectory_subset_dict
) = load_pickle_bundle(os.path.join(temp_move, "_experiment_development_atlas_helper_pickle_bundle.pkl"))
ground_truth_trajectories.keys()
# %%
InputTrajectories(ground_truth_trajectories).get_trajectory("germline", False)
# %%
InputTrajectories(ground_truth_trajectories).get_trajectory("Haem_ref", False)
# %%
InputTrajectories(ground_truth_trajectories).get_trajectory("pns_neuro", False)
# %%
fig, ax = InputTrajectories(ground_truth_trajectories).get_trajectory("germline", False).plot_trajectory(
title="",
figsize=(12, 1.65),
label_offset=0,
return_ax=True
)
plt.savefig("/Users/kemalinecik/Downloads/fig2fig/graph_germline.pdf")
# %%
fig, ax = InputTrajectories(ground_truth_trajectories).get_trajectory("Haem_ref", False).plot_trajectory(
title="",
figsize=(12, 7),
label_offset=0,
return_ax=True
)
plt.savefig("/Users/kemalinecik/Downloads/fig2fig/graph_Haem_ref.pdf")
# %%
fig, ax = InputTrajectories(ground_truth_trajectories).get_trajectory("pns_neuro", False).plot_trajectory(
title="",
figsize=(12, 2),
label_offset=0,
return_ax=True
)
plt.savefig("/Users/kemalinecik/Downloads/fig2fig/graph_pns_neuro.pdf")
# %% [markdown]
# # Figure
# %%
def logistic_invert_columns(df, col=None, k=1.0):
"""
Apply logistic inversion scaling to a column or entire DataFrame.
Parameters
----------
df : pd.DataFrame or pd.Series
Input DataFrame or Series.
col : str or None, default=None
Column name to transform. If None, apply to all numeric columns.
k : float, default=1.0
Steepness of the sigmoid. Larger k -> sharper curve.
Returns
-------
pd.DataFrame or pd.Series
Transformed DataFrame/Series with values in (0, 1).
"""
if isinstance(df, pd.Series): # if a single column (Series)
return 1 / (1 + np.exp(k * (df - df.mean())))
if col is not None: # single column of DataFrame
df = df.copy()
x = df[col]
df[col] = 1 / (1 + np.exp(k * (x - x.mean())))
return df
# all numeric columns
df = df.copy()
for c in df.select_dtypes(include=np.number).columns:
x = df[c]
df[c] = 1 / (1 + np.exp(k * (x - x.mean())))
return df
def logistic_invert_rowwise(df, k=1.0):
"""
Apply logistic inversion scaling row-wise to all numeric columns.
Parameters
----------
df : pd.DataFrame
Input DataFrame.
k : float, default=1.0
Steepness of the sigmoid. Larger k -> sharper curve.
Returns
-------
pd.DataFrame
Transformed DataFrame with values in (0, 1).
"""
df = df.copy()
numeric_cols = df.select_dtypes(include=np.number).columns
# Apply rowwise
def transform_row(row):
mean_val = row[numeric_cols].mean()
return 1 / (1 + np.exp(k * (row[numeric_cols] - mean_val)))
df[numeric_cols] = df[numeric_cols].apply(transform_row, axis=1)
return df
# %%
def rank_metrics(df, metric_direction_dict, metric_index_pos, method='min'):
ranked_df = pd.DataFrame(index=df.index, columns=df.columns, dtype=float)
for indices, row in df.iterrows():
metric = indices[metric_index_pos]
direction = metric_direction_dict.get(metric)
if direction not in ('increasing', 'decreasing'):
raise ValueError
numeric_row = pd.to_numeric(row, errors='coerce')
if numeric_row.isna().any():
ranked_df.loc[indices] = np.nan
continue
if direction == 'increasing':
ranks = numeric_row.rank(method='dense', ascending=False)
else:
ranks = numeric_row.rank(method='dense', ascending=True)
# ranks = 1 - pd.Series(MinMaxScaler().fit_transform(row.values.reshape(-1, 1)).flatten(), index=row.index)
ranked_df.loc[indices] = ranks.astype('Float64')
return ranked_df
result_df = pd.read_pickle(os.path.join(temp_move, "_experiment_development_atlas_result_df_dataframe.pkl"))
df_pivot = result_df.pivot(index=["trajectory_class", 'trajectory', 'path', 'metric'], columns='representation', values='score')
df_pivot_rank = rank_metrics(df_pivot, metric_direction_dict, metric_index_pos=3, method='min')
df_pivot_rank_mean = df_pivot_rank.groupby(level=1).mean()
rename_model_name_dict = {
'pca': 'PCA',
'X_pca': 'PCA',
'harmony': 'Harmony',
'invae': 'inVAE',
'scanvi': 'scANVI',
'scvi': 'scVI',
'scpoli': 'scPoli',
'scanoroma': 'Scanoroma',
'tardis_1': 'DELETE ME',
'tardis_2': 'TarDis',
'tardis': 'TarDis',
}
col_order = ['Scanoroma', 'Harmony', 'TarDis', 'PCA', 'scPoli', 'scVI']
sctram_summary = (logistic_invert_rowwise(df_pivot_rank).groupby(level=[2, 3]).mean().groupby(level=0).mean())
_sctram_summary = (pd.DataFrame(logistic_invert_rowwise(df_pivot_rank).groupby(level=3).mean().mean(), columns=["Total scTRAM"]).T)
sctram_summary = pd.concat([sctram_summary, _sctram_summary], axis=0)
sctram_summary = sctram_summary.rename(
index={"adjacency": "Adjacency-based", "embedding": "Embedding-based", "pseudotime": "Pseudotime-based"},
columns=rename_model_name_dict,
)[col_order]
sctram_summary.index.name = None
sctram_summary.columns.name = "Representation"
sctram_summary
# %%
# sctram_summary = pd.read_pickle(os.path.join(temp_move, "_experiment_development_atlas_sctram_summary_dataframe.pkl")) # invert back to rank again
scib_summary = pd.read_pickle(os.path.join(temp_move, "_experiment_development_atlas_scib_summary_dataframe.pkl"))
replacements = {
"Pseudotime-based": "Cell Order\ncontinuity",
"Embedding-based": "Trajectory\ndominance",
"Adjacency-based": "Topology\nconservation",
"Bio conservation": "Biological\nconservation",
"Batch correction": "Batch\ncorrection",
}
summary = pd.concat([scib_summary, sctram_summary], keys=["scIB", "scTRAM"], axis=0)
summary.index.names = ["Source", "Metric Group"]
summary.columns.name = "Representation"
new_index_tuples = [
("Aggregate score", metric.split("Total ")[1]) if metric.startswith("Total ") else (src, metric) for src, metric in summary.index
]
summary.index = pd.MultiIndex.from_tuples(new_index_tuples, names=summary.index.names)
weights = { # just provisional for now
"scTRAM": 0.25,
"scIB": 0.75,
}
row_ts = summary.loc[("Aggregate score", "scTRAM")]
row_si = summary.loc[("Aggregate score", "scIB")]
combined = weights["scTRAM"] * row_ts + weights["scIB"] * row_si
combined_df = pd.DataFrame(combined).T
combined_df.index = pd.MultiIndex.from_tuples([("Aggregate score", "Total")], names=summary.index.names)
summary = pd.concat([summary, combined_df])
summary = summary.sort_index(ascending=False)
summary = summary.sort_values(by=[("Aggregate score", "Total")], axis=1, ascending=False).astype(np.float64)
summary.rename(index=replacements, level="Metric Group", inplace=True)
summary_T = summary.T
summary
# %%
df_plot = summary_T.reset_index().copy()
df_plot_dict = {f"{lvl0}\n{lvl1}": (lvl0, lvl1) for (lvl0, lvl1) in df_plot.columns}
df_plot.columns = [f"{lvl0}\n{lvl1}" for (lvl0, lvl1) in df_plot.columns]
cmap_fn = lambda col_data: normed_cmap(col_data, cmap=cm.PRGn, num_stds=2.5)
col_defs = []
for col in df_plot.columns:
# Create a “normalized color map” for this column:
lvl0, lvl1 = df_plot_dict[col]
if lvl0 == "Aggregate score" or lvl0 == "Representation":
continue
cmap = cmap_fn(df_plot[col])
cd = ColumnDefinition(
name=col, # Must exactly match the flattened column name in df_plot
width=0.4, # width of each column in inches (tweak as needed)
cmap=cmap, # cell background colormap (normed by this column’s values)
text_cmap=cmap, # text color also follows same colormap
textprops={
"ha": "center",
"bbox": {"boxstyle": "circle", "pad": 0},
},
group=lvl0,
title=lvl1,
formatter="{:.3f}".format, # three decimal places
)
col_defs.append(cd)
i = 0
for col in df_plot.columns:
# Create a “normalized color map” for this column:
lvl0, lvl1 = df_plot_dict[col]
if lvl0 != "Aggregate score" or lvl0 == "Representation":
continue
cmap = cmap_fn(df_plot[col])
i += 1
c_min = min(df_plot[col])
c_max = max(df_plot[col])
cmap = cmap_fn(df_plot[col])
cd = ColumnDefinition(
name=col,
width=0.3,
plot_fn=bar,
plot_kw={
"cmap": cmap, #cm.YlGnBu,
"plot_bg_bar": False,
# "xlim": [c_min, c_max],
"annotate": True,
"height": 0.7,
"formatter": "{:.3f}",
"textprops": {"fontsize": 10}
},
group=lvl0,
title=lvl1,
border="left" if i == 0 else None,
)
col_defs.append(cd)
col_defs += [ColumnDefinition(name="Representation\n", title="", width=0.3, textprops={"ha": "left", "weight": "bold"})]
# Allow to manipulate text post-hoc (in illustrator)
with mpl.rc_context({"svg.fonttype": "none"}):
fig, ax = plt.subplots(figsize=(len(summary_T.columns) * 1.3, 3 + 0.25 * len(summary_T.index)))
tab = Table(
df_plot,
cell_kw={
"linewidth": 0,
"edgecolor": "k",
},
row_dividers=True,
footer_divider=True,
textprops={"fontsize": 10, "ha": "center"},
row_divider_kw={"linewidth": 0.25, "linestyle": "-"},
col_label_divider_kw={"linewidth": 1, "linestyle": "-"},
column_border_kw={"linewidth": 1, "linestyle": "-"},
column_definitions=col_defs,
even_row_color="#97ffff",
odd_row_color="#ffffff",
ax=ax,
index_col="Representation\n",
).autoset_fontcolors(colnames=df_plot.columns)
plt.savefig("/Users/kemalinecik/Downloads/fig2fig/table.pdf")
# %% [markdown]
# # Figure
# %%
result_df = pd.read_pickle(os.path.join(temp_move, "_experiment_development_atlas_result_df_dataframe.pkl"))
df_pivot = result_df.pivot(index=["trajectory_class", 'trajectory', 'path', 'metric'], columns='representation', values='score')
df_pivot_rank = rank_metrics(df_pivot, metric_direction_dict, metric_index_pos=3, method='min')
df_pivot_rank_mean = logistic_invert_rowwise(df_pivot_rank).groupby(level=1).mean()
rename_model_name_dict = {
'pca': 'PCA',
'X_pca': 'PCA',
'harmony': 'Harmony',
'invae': 'inVAE',
'scanvi': 'scANVI',
'scvi': 'scVI',
'scpoli': 'scPoli',
'scanoroma': 'Scanoroma',
'tardis_1': 'DELETE ME',
'tardis_2': 'TarDis',
'tardis': 'TarDis',
}
col_order = ['Scanoroma', 'Harmony', 'TarDis', 'PCA', 'scPoli', 'scVI']
def _(df):
df = df.rename(columns=rename_model_name_dict)
df = df[col_order]
fig_height=4
n_cols = df.shape[1]
col_width = 0.7 # Width per column in inches
fig_width = n_cols * col_width
# fig_height = 8 # Fixed height for better vertical readability
# Create figure with dynamic width
plt.figure(figsize=(fig_width, fig_height))
# Create boxplot with refined styling
bp = sns.boxplot(data=df,
width=0.8,
showfliers=False,
boxprops=dict(facecolor='lightgray', edgecolor='black', linewidth=0.81),
whiskerprops=dict(color='black', linewidth=0.81),
medianprops=dict(color='darkred', linewidth=0.81),
capprops=dict(color='black', linewidth=0.81),
saturation=1,
zorder=1)
# Add swarmplot with sophisticated aesthetics
sp = sns.swarmplot(data=df,
size=4,
palette='dark:black',
edgecolor='white',
linewidth=0.6,
alpha=0.8,
marker='o',
zorder=2)
# Enhance axis styling
plt.xticks(rotation=90, ha='center', fontsize=10)
plt.yticks(fontsize=10)
plt.xlabel('Representation', fontsize=10, labelpad=10)
plt.ylabel("Performance", fontsize=10, labelpad=10)
# plt.gca().set_ylim(1.75, 5.5)
plt.gca().yaxis.set_major_formatter(FormatStrFormatter('%.2f'))
# Add light grid and adjust borders
bp.grid(axis='y', linestyle='--', alpha=0.4)
sns.despine(left=True, trim=False)
# Adjust layout with professional spacing
plt.tight_layout(rect=[0.05, 0.05, 0.95, 0.95])
plt.savefig("/Users/kemalinecik/Downloads/fig2fig/modelcomp_boxplot.pdf")
_(df_pivot_rank_mean)
# %%
df_pivot_rank_mean = logistic_invert_rowwise(df_pivot_rank).groupby(level=[0, 1]).mean()
df_pivot_rank_mean = df_pivot_rank_mean.rename(columns=rename_model_name_dict)
df_pivot_rank_mean = df_pivot_rank_mean[col_order]
df_long = df_pivot_rank_mean.reset_index().melt(id_vars=['trajectory_class', 'trajectory'], var_name='representation', value_name='score')
# %%
# match boxplot’s height & per-column width
fig_height = 3
n_cols = df_long["representation"].nunique()
col_width = 1.4
fig_width = n_cols * col_width
n_tc = len(np.unique(df_long["trajectory_class"]))
fig, ax = plt.subplots(figsize=(fig_width, fig_height))
# draw bars
bp = sns.barplot(
data=df_long,
x="representation",
y="score",
hue="trajectory_class",
palette=sns.color_palette("Greys", n_colors=n_tc+2),
width=0.8,
saturation=1,
errorbar="sd",
capsize=0.05,
err_kws={'color': 'black', 'linewidth': 0.81},
edgecolor="black",
linewidth=0.81,
zorder=1,
ax=ax
)
# enforce edge styling
for patch in bp.patches:
patch.set_edgecolor("black")
patch.set_linewidth(0.81)
# axes
ax.set_xticklabels(ax.get_xticklabels(), rotation=90, ha='center', fontsize=10)
ax.tick_params(axis="y", labelsize=10)
ax.set_xlabel('Representation', fontsize=10, labelpad=10)
plt.ylabel("Performance", fontsize=10, labelpad=10)
ax.set_ylim(0.1, None)
ax.yaxis.set_major_formatter(FormatStrFormatter('%.2f'))
# grid + despine
ax.grid(axis='y', linestyle='--', alpha=0.4)
sns.despine(left=True, trim=False)
# remove the old legend
ax.legend_.remove()
# build new, professional legend
handles, labels = ax.get_legend_handles_labels()
slim_handles = [
mpatches.Patch(
facecolor=h.get_facecolor(), # <-- keep the whole RGBA tuple
edgecolor="black",
linewidth=0.5 # thinner border
)
for h in handles
]
leg = ax.legend(
slim_handles,
labels,
title="Trajectory Group",
loc="center left",
bbox_to_anchor=(1.02, 0.5),
frameon=False,
alignment="left",
title_fontsize=8,
fontsize=8,
borderaxespad=0.5,
handletextpad=0.4,
labelspacing=0.5,
)
# tighten layout (accounts for legend)
plt.tight_layout(rect=[0, 0, 0.85, 1])
plt.savefig("/Users/kemalinecik/Downloads/fig2fig/modelcomp_barplot1.pdf")
# %%
df_pivot_rank_mean = logistic_invert_rowwise(df_pivot_rank).groupby(level=[2, 1]).mean()
df_pivot_rank_mean = df_pivot_rank_mean.rename(columns=rename_model_name_dict)
df_pivot_rank_mean = df_pivot_rank_mean[col_order]
df_long = df_pivot_rank_mean.reset_index().melt(id_vars=['path', 'trajectory'], var_name='representation', value_name='score')
# %%
label_mapping = {
"adjacency": 'Topology conservation',
"embedding": 'Trajectory dominance',
"pseudotime": 'Cell Order continuity',
}
# match boxplot’s height & per-column width
fig_height = 3
n_cols = df_long["representation"].nunique()
col_width = 1.4
fig_width = n_cols * col_width
fig, ax = plt.subplots(figsize=(fig_width, fig_height))
# draw bars
bp = sns.barplot(
data=df_long,
x="representation",
y="score",
hue="path",
palette=sns.color_palette("Greys", n_colors=len(label_mapping)+2),
width=0.8,
saturation=1,
errorbar="sd",
capsize=0.05,
err_kws={'color': 'black', 'linewidth': 0.81},
edgecolor="black",
linewidth=0.81,
zorder=1,
ax=ax
)
# enforce edge styling
for patch in bp.patches:
patch.set_edgecolor("black")
patch.set_linewidth(0.81)
# axes
ax.set_xticklabels(ax.get_xticklabels(), rotation=90, ha='center', fontsize=10)
ax.tick_params(axis="y", labelsize=10)
ax.set_xlabel('Representation', fontsize=10, labelpad=10)
plt.ylabel("Performance", fontsize=10, labelpad=10)
ax.set_ylim(0.1, None)
ax.yaxis.set_major_formatter(FormatStrFormatter('%.2f'))
# grid + despine
ax.grid(axis='y', linestyle='--', alpha=0.4)
sns.despine(left=True, trim=False)
# remove the old legend
ax.legend_.remove()
# build new, professional legend
handles, labels = ax.get_legend_handles_labels()
new_labels = [label_mapping.get(l, l) for l in labels]
slim_handles = [
mpatches.Patch(
facecolor=h.get_facecolor(), # <-- keep the whole RGBA tuple
edgecolor="black",
linewidth=0.5 # thinner border
)
for h in handles
]
leg = ax.legend(
slim_handles,
new_labels,
title="Metric Group",
loc="center left",
bbox_to_anchor=(1.02, 0.5),
alignment="left",
frameon=False,
title_fontsize=8,
fontsize=8,
borderaxespad=0.5,
handletextpad=0.4,
labelspacing=0.5,
)
# tighten layout (accounts for legend)
plt.tight_layout(rect=[0, 0, 0.85, 1])
plt.savefig("/Users/kemalinecik/Downloads/fig2fig/modelcomp_barplot2.pdf")
# %% [markdown]
# # Figure 2
# %%
result_df_main = pd.read_pickle(os.path.join(temp_move, f"df_boots_fig2_calculations.pkl"))
(
df_hdca_component_adata_paths,
df_hdca_lvl3_adata_paths,
out_df,
ground_truth_trajectories,
ground_truth_trajectories_annotation_level,
trajectory_subset_dict
) = load_pickle_bundle(os.path.join(temp_move, "_experiment_development_atlas_helper_pickle_bundle.pkl"))
# %%
from math import sqrt
def rank_metrics(df, metric_direction_dict, metric_index_pos, method='min'):
ranked_df = pd.DataFrame(index=df.index, columns=df.columns, dtype=float)
for indices, row in df.iterrows():
metric = indices[metric_index_pos]
direction = metric_direction_dict.get(metric)
if direction not in ('increasing', 'decreasing'):
raise ValueError
numeric_row = pd.to_numeric(row, errors='coerce')
if numeric_row.isna().any():
ranked_df.loc[indices] = np.nan
continue
if direction == 'increasing':
ranks = numeric_row.rank(method='dense', ascending=False)
else:
ranks = numeric_row.rank(method='dense', ascending=True)
# ranks = 1 - pd.Series(MinMaxScaler().fit_transform(row.values.reshape(-1, 1)).flatten(), index=row.index)
ranked_df.loc[indices] = ranks.astype('Float64')
return ranked_df
def plot_trajectory(
G,
title="trajectory_name",
figsize=None,
font_size=10,
node_size=300,
node_color="#8DA0CB", # Modern muted blue
cmap="viridis",
color_nodes_by_position=False,
# Edge styling parameters
edge_width=1,
edge_color="gray",
color_edges_by_weight=False,
edge_weight_attribute="weight",
edge_cmap="plasma",
edge_vmin=None,
edge_vmax=None,
edge_colorbar=True,
edge_labels=False,
edge_label_format=".2f",
edge_label_font_size=8,
edge_label_color="black",
edge_label_offset=None,
# Layout parameters
label_offset=None,
title_y=1.02,
layout_args="-Grankdir=LR -Gnodesep=0.25 -Granksep=0.25",
file_name="blabla"
):
pos = nx.nx_agraph.graphviz_layout(G, prog="dot", args=layout_args)
x_coords = [pos[node][0] for node in G.nodes()] if G.nodes() else []
y_coords = [pos[node][1] for node in G.nodes()] if G.nodes() else []
# Create actual figure with determined size
fig = plt.figure(figsize=figsize)
ax = fig.add_subplot(111)
# Node coloring logic
if color_nodes_by_position:
try:
min_x, max_x = min(x_coords), max(x_coords)
color_vals = [(x - min_x) / (max_x - min_x) if max_x != min_x else 0.5 for x in x_coords]
node_color = plt.get_cmap(cmap)(color_vals)
except Exception as e:
print(f"Warning: Positional node coloring failed - {e}. Using base color.")
# Edge coloring logic
if color_edges_by_weight:
weights = [G[u][v].get(edge_weight_attribute, 1.0) for u, v in G.edges]
# Separate NaN and valid weights
valid_weights = [w for w in weights if not np.isnan(w)]
if valid_weights:
edge_vmin = edge_vmin if edge_vmin is not None else min(valid_weights)
edge_vmax = edge_vmax if edge_vmax is not None else max(valid_weights)
norm = mpl.colors.Normalize(vmin=edge_vmin, vmax=edge_vmax)
cmap_edge = plt.get_cmap(edge_cmap)
# Create edge colors, using gray with alpha=0.712 for NaN weights
edge_colors = []
edge_alphas = []
for w in weights:
if np.isnan(w):
edge_colors.append("lightgray")
edge_alphas.append(0.712)
else:
edge_colors.append(cmap_edge(norm(w)))
edge_alphas.append(1.0)
sm = plt.cm.ScalarMappable(norm=norm, cmap=cmap_edge)
sm.set_array(valid_weights)
else:
edge_colors = "lightgray"
edge_alphas = 0.712
sm = None
else:
edge_colors = edge_color
edge_alphas = 1.0
sm = None
# Label positioning offsets
if label_offset is None:
label_offset = 0
if edge_label_offset is None:
edge_label_offset = 0
# Calculate final label positions
label_pos = {n: (x, y + label_offset) for n, (x, y) in pos.items()}
# Draw nodes and edges
nx.draw_networkx_nodes(
G, pos, ax=ax, node_size=node_size, node_color=node_color, edgecolors="black", linewidths=0.8, alpha=1.0
)
node_radius = sqrt(node_size) / 2.0
if color_edges_by_weight and isinstance(edge_alphas, list):
# Draw edges individually when we have different alpha values
for i, (u, v) in enumerate(G.edges()):
nx.draw_networkx_edges(
G,
pos,
edgelist=[(u, v)],
ax=ax,
edge_color=[edge_colors[i]],
width=edge_width,
alpha=edge_alphas[i],
connectionstyle="arc3",
arrows=False,
)
else:
nx.draw_networkx_edges(
G,
pos,
ax=ax,
edge_color=edge_colors,
width=edge_width,
alpha=edge_alphas if not isinstance(edge_alphas, list) else 1.0,
connectionstyle="arc3",
arrows=False,
)
# Edge label placement
if edge_labels:
for u, v in G.edges():
if edge_weight_attribute not in G[u][v]:
continue
weight = G[u][v][edge_weight_attribute]
x1, y1 = pos[u]
x2, y2 = pos[v]
dx, dy = x2 - x1, y2 - y1
length = (dx**2 + dy**2) ** 0.5
if length == 0:
continue # Skip self-loops
# Calculate perpendicular offset position
mid_x = (x1 + x2) / 2 + (dy / length) * edge_label_offset
mid_y = (y1 + y2) / 2 - (dx / length) * edge_label_offset
ax.text(
mid_x,
mid_y,
f"{weight:{edge_label_format}}",
fontsize=edge_label_font_size,
color=edge_label_color,
ha="center",
va="center",
# bbox=dict(boxstyle="round", facecolor="white", alpha=0.8, edgecolor="none"),
)
# Edge weight colorbar
if color_edges_by_weight and edge_colorbar and sm:
cbar = plt.colorbar(sm, ax=ax, orientation="horizontal", shrink=0.3, aspect=40, pad=0.05)
cbar.set_label("Edge Weight", fontsize=font_size)
cbar.ax.tick_params(labelsize=font_size - 2)
fig.suptitle(title, y=title_y, fontsize=font_size + 2, fontweight="bold", color="#333333")
# Final layout adjustments
plt.subplots_adjust(
left=0.03, right=0.97, top=0.97, bottom=0.15 if (color_edges_by_weight and edge_colorbar) else 0.03
)
ax.set_axis_off()
plt.margins(y=1) # Add 20% margin on all sides
plt.savefig(f"/Users/kemalinecik/Downloads/fig2fig/{file_name}.pdf")
# %%
for trajectory_focus, result_df in result_df_main.groupby('trajectory'):
if trajectory_focus == "Haem_ref":
decompose_method = "2"
input_trajectories_path_litc = os.path.join(temp_move, f"adata_hdca_input_{trajectory_focus}_litc_{decompose_method}.pkl")
with open(input_trajectories_path_litc, 'rb') as _file:
lit = pickle.load(_file)
break
df_edges = []
for ind, row in result_df.iterrows():
trj = lit.get_trajectory(row["booted_trajectory"], False)
for from_cell, to_cell in trj.edges():
row_copy = row.copy()
row_copy["from_cell"] = from_cell
row_copy["to_cell"] = to_cell
df_edges.append(row_copy)
df_edges = pd.DataFrame(df_edges)
df_edges.reset_index(drop=True, inplace=True)
# %%
df_edges_pivot = df_edges.pivot(index=['path', 'booted_trajectory', 'trajectory', 'metric', 'from_cell', 'to_cell'], columns='representation', values='score')
df_edges_pivot_rank = rank_metrics(df_edges_pivot, metric_direction_dict, metric_index_pos=3, method='min')
df_edges_pivot_rank_mean = logistic_invert_rowwise(df_edges_pivot_rank).groupby(level=[4, 5]).mean()
# %%
from scipy.stats import mannwhitneyu, binomtest, wilcoxon
def compare_scores_and_get_means(group):
scpoli_scores = group['scpoli'].dropna()
scvi_scores = group['scvi'].dropna()
if len(scpoli_scores) < 2 or len(scvi_scores) < 2:
raise ValueError
# since your “scores” are actually ranks (ordinal) for the same boot replicate per edge, you have paired, ordinal data.
# Perform the Mann-Whitney U test
# stat, p_value = mannwhitneyu(scpoli_scores, scvi_scores, alternative='two-sided')
# Wilcoxon signed-rank — more power if the distribution of paired differences is roughly symmetric; handle ties with zero_method="pratt"
method = 'exact' if len(scpoli_scores) <= 25 else 'approx'
p_value = wilcoxon(scpoli_scores, scvi_scores, zero_method='pratt', alternative='two-sided', method=method).pvalue
# Statistically different: return means.
return pd.Series({
'mean_scpoli': np.mean(scpoli_scores),
'mean_scvi': np.mean(scvi_scores),
'p_value': p_value
})
df_edges_pivot_rank_p = logistic_invert_rowwise(df_edges_pivot_rank).groupby(['from_cell', 'to_cell']).apply(compare_scores_and_get_means).reset_index()
alpha = 0.05 / len(df_edges_pivot_rank_p)
df_edges_pivot_rank_p['mean_scvi_final'] = np.where(df_edges_pivot_rank_p['p_value'] < alpha, df_edges_pivot_rank_p['mean_scvi'], np.nan)
df_edges_pivot_rank_p['mean_scpoli_final'] = np.where(df_edges_pivot_rank_p['p_value'] < alpha, df_edges_pivot_rank_p['mean_scpoli'], np.nan)
df_edges_pivot_rank_p
# %%
trj = InputTrajectories(ground_truth_trajectories).get_trajectory(trajectory_focus, False)
representations = ["scvi", "scpoli"]
for ind, row in df_edges_pivot_rank_p.iterrows():
for repr in representations:
trj[row["from_cell"]][row["to_cell"]][f"weight_{repr}"] = row[f"mean_{repr}_final"]
# %%
for i in ['weight_scpoli', 'weight_scvi']:
plot_trajectory(
trj,
figsize=(4, 11),
edge_weight_attribute=i,
title="",
edge_vmin=0.4,
edge_vmax=0.7,
color_edges_by_weight=True,
node_size=150,
node_color="lightgray",
edge_color='black',
edge_width=5,
edge_labels=False,
edge_label_font_size=5,
edge_cmap='viridis',
edge_colorbar=False,
file_name=f"{trajectory_focus}_{i}"
)
# %%
for trajectory_focus, result_df in result_df_main.groupby('trajectory'):
if trajectory_focus == "germline":
decompose_method = "2"
input_trajectories_path_litc = os.path.join(temp_move, f"adata_hdca_input_{trajectory_focus}_litc_{decompose_method}.pkl")
with open(input_trajectories_path_litc, 'rb') as _file:
lit = pickle.load(_file)
break
df_edges = []
for ind, row in result_df.iterrows():
trj = lit.get_trajectory(row["booted_trajectory"], False)
for from_cell, to_cell in trj.edges():
row_copy = row.copy()
row_copy["from_cell"] = from_cell
row_copy["to_cell"] = to_cell
df_edges.append(row_copy)
df_edges = pd.DataFrame(df_edges)
df_edges.reset_index(drop=True, inplace=True)
df_edges_pivot = df_edges.pivot(index=['path', 'booted_trajectory', 'trajectory', 'metric', 'from_cell', 'to_cell'], columns='representation', values='score')
df_edges_pivot_rank = rank_metrics(df_edges_pivot, metric_direction_dict, metric_index_pos=3, method='min')
df_edges_pivot_rank_mean = logistic_invert_rowwise(df_edges_pivot_rank).groupby(level=[4, 5]).mean()
df_edges_pivot_rank_p = logistic_invert_rowwise(df_edges_pivot_rank).groupby(['from_cell', 'to_cell']).apply(compare_scores_and_get_means).reset_index()
alpha = 0.05 / len(df_edges_pivot_rank_p)
df_edges_pivot_rank_p['mean_scvi_final'] = np.where(df_edges_pivot_rank_p['p_value'] < alpha, df_edges_pivot_rank_p['mean_scvi'], np.nan)
df_edges_pivot_rank_p['mean_scpoli_final'] = np.where(df_edges_pivot_rank_p['p_value'] < alpha, df_edges_pivot_rank_p['mean_scpoli'], np.nan)
trj = InputTrajectories(ground_truth_trajectories).get_trajectory(trajectory_focus, False)
representations = ["scvi", "scpoli"]
for ind, row in df_edges_pivot_rank_p.iterrows():
for repr in representations:
trj[row["from_cell"]][row["to_cell"]][f"weight_{repr}"] = row[f"mean_{repr}_final"]
# %%
for i in ['weight_scpoli', 'weight_scvi']:
plot_trajectory(
trj,
figsize=(4, 0.75),
edge_weight_attribute=i,
title="",
edge_vmin=0.4,
edge_vmax=0.7,
color_edges_by_weight=True,
node_size=150,
node_color="lightgray",
edge_color='black',
edge_width=5,
edge_labels=False,
edge_label_font_size=5,
edge_cmap='viridis',
edge_colorbar=False,
file_name=f"{trajectory_focus}_{i}"
)
# %%
def compare_scores_and_get_means(group):
tardis_scores = group['tardis'].dropna()
scvi_scores = group['scvi'].dropna()
if len(tardis_scores) < 2 or len(scvi_scores) < 2:
raise ValueError
# since your “scores” are actually ranks (ordinal) for the same boot replicate per edge, you have paired, ordinal data.
# Perform the Mann-Whitney U test
# stat, p_value = mannwhitneyu(tardis_scores, scvi_scores, alternative='two-sided')
# Wilcoxon signed-rank — more power if the distribution of paired differences is roughly symmetric; handle ties with zero_method="pratt"
method = 'exact' if len(tardis_scores) <= 25 else 'approx'
p_value = wilcoxon(tardis_scores, scvi_scores, zero_method='pratt', alternative='two-sided', method=method).pvalue
# Statistically different: return means.
return pd.Series({
'mean_tardis': np.mean(tardis_scores),
'mean_scvi': np.mean(scvi_scores),
'p_value': p_value
})
# %%
for trajectory_focus, result_df in result_df_main.groupby('trajectory'):
if trajectory_focus == "pns_neuro":
decompose_method = "2"
input_trajectories_path_litc = os.path.join(temp_move, f"adata_hdca_input_{trajectory_focus}_litc_{decompose_method}.pkl")
with open(input_trajectories_path_litc, 'rb') as _file:
lit = pickle.load(_file)
break
df_edges = []
for ind, row in result_df.iterrows():
trj = lit.get_trajectory(row["booted_trajectory"], False)
for from_cell, to_cell in trj.edges():
row_copy = row.copy()
row_copy["from_cell"] = from_cell
row_copy["to_cell"] = to_cell
df_edges.append(row_copy)
df_edges = pd.DataFrame(df_edges)
df_edges.reset_index(drop=True, inplace=True)
df_edges_pivot = df_edges.pivot(index=['path', 'booted_trajectory', 'trajectory', 'metric', 'from_cell', 'to_cell'], columns='representation', values='score')
df_edges_pivot_rank = rank_metrics(df_edges_pivot, metric_direction_dict, metric_index_pos=3, method='min')
df_edges_pivot_rank_mean = logistic_invert_rowwise(df_edges_pivot_rank).groupby(level=[4, 5]).mean()
df_edges_pivot_rank_p = logistic_invert_rowwise(df_edges_pivot_rank).groupby(['from_cell', 'to_cell']).apply(compare_scores_and_get_means).reset_index()
alpha = 0.05 / len(df_edges_pivot_rank_p)
df_edges_pivot_rank_p['mean_scvi_final'] = np.where(df_edges_pivot_rank_p['p_value'] < alpha, df_edges_pivot_rank_p['mean_scvi'], np.nan)
df_edges_pivot_rank_p['mean_tardis_final'] = np.where(df_edges_pivot_rank_p['p_value'] < alpha, df_edges_pivot_rank_p['mean_tardis'], np.nan)
trj = InputTrajectories(ground_truth_trajectories).get_trajectory(trajectory_focus, False)
representations = ["scvi", "tardis"]
for ind, row in df_edges_pivot_rank_p.iterrows():
for repr in representations:
trj[row["from_cell"]][row["to_cell"]][f"weight_{repr}"] = row[f"mean_{repr}_final"]
# %%
for i in ['weight_tardis', 'weight_scvi']:
plot_trajectory(
trj,
figsize=(4, 3.25),
edge_weight_attribute=i,
title="",
edge_vmin=0.4,
edge_vmax=0.7,
color_edges_by_weight=True,
node_size=150,
node_color="lightgray",
edge_color='black',
edge_width=5,
edge_labels=False,
edge_label_font_size=5,
edge_cmap='viridis',
edge_colorbar=False,
file_name=f"{trajectory_focus}_{i}"
)
# %%
from matplotlib import ticker
def create_vertical_colorbar(
label,
filename,
vmin=2.5,
vmax=3.5,
cmap='plasma',
fontsize=10,
n_ticks=5,
fig_height=4,
fig_width=1,
scale=1.0
):
"""
Creates and displays a standalone vertical colorbar.
Parameters:
- label: str, label for the colorbar
- vmin, vmax: float, data range
- cmap: str, colormap name
- fontsize: int, for label & tick labels
- n_ticks: int, approx number of ticks to show
- fig_height: float, base height in inches
- fig_width: float, base width in inches
- scale: float, scale factor for overall size
"""
# 1) Make the ScalarMappable
norm = mpl.colors.Normalize(vmin=vmin, vmax=vmax)
sm = mpl.cm.ScalarMappable(norm=norm, cmap=cmap)
sm.set_array([])
# 2) New figure, scaled
fig = plt.figure(figsize=(fig_width * scale, fig_height * scale))
# you can tweak these numbers if you want more/less padding
cax = fig.add_axes([0.2, 0.05, 0.1, 0.9])
# 3) Draw the colorbar
cbar = fig.colorbar(
sm,
cax=cax,
orientation='vertical'
)
# 4) Automatic “nice” locators + re‐draw
cbar.locator = ticker.MaxNLocator(nbins=n_ticks)
cbar.update_ticks()
# 5) Force one‐decimal formatting
cbar.ax.yaxis.set_major_formatter(ticker.FormatStrFormatter('%.1f'))
# 6) Label & style
cbar.set_label(label, fontsize=fontsize)
cbar.ax.tick_params(labelsize=fontsize)
plt.subplots_adjust(left=0.03, right=0.97, top=0.97, bottom=0.15)
plt.savefig(f"/Users/kemalinecik/Downloads/fig2fig/{filename}.pdf")
create_vertical_colorbar("Performance", filename = "colorbar_performance_rank",
cmap="viridis", n_ticks=3, fig_height=2, fig_width=2.4, scale=0.8, fontsize=10, vmin=0.4, vmax=0.7)
# %% [markdown]
# # Figure 3
# %%
result_df_boot = pd.read_pickle(os.path.join(temp_move, f"df_boots_fig2_calculations.pkl"))
result_df_boot = result_df_boot[result_df_boot["representation"]=="scvi"]
result_df_boot.reset_index(inplace=True, drop=True)
result_df_main = pd.read_pickle(os.path.join(temp_move, f"df_boots_fig2_calculations_for_individuals.pkl"))
# %%
focus_trajectory = "germline"
df_compo = result_df_main[result_df_main["trajectory"] == focus_trajectory]
df_atlas = result_df_boot[result_df_boot["trajectory"] == focus_trajectory]
df_atlas.reset_index(inplace=True, drop=True)
df_compo.reset_index(inplace=True, drop=True)
decompose_method = "2"
input_trajectories_path_litc = os.path.join(temp_move, f"adata_hdca_input_{focus_trajectory}_litc_{decompose_method}.pkl")
with open(input_trajectories_path_litc, 'rb') as _file:
lit = pickle.load(_file)
trj_size_dict = dict()
for trj_name in lit.graph['trajectories']:
trj_size_dict[trj_name] = len(lit.get_trajectory(trj_name, False).nodes()), len(lit.get_trajectory(trj_name, False).edges())
assert df_compo.shape == df_atlas.shape
for i in range(len(df_compo)):
i1 = list(df_atlas.iloc[i][["booted_trajectory", "metric"]])
i2 = list(df_compo.iloc[i][["booted_trajectory", "metric"]])
assert i1==i2
# df_both
df_both = df_compo.copy()
del df_both["score"]
df_both["score_Component"] = list(df_compo["score"])
df_both["score_Atlas"] = list(df_atlas["score"])
df_both.reset_index(inplace=True, drop=True)
del df_both["anndata_subset"]
def get_winners(row):
metric = row['metric']
direction = metric_direction_dict[metric]
c = row['score_Component']
a = row['score_Atlas']
if direction == 'decreasing':
if c > a:
return pd.Series([0, 1])
elif a > c:
return pd.Series([1, 0])
else:
return pd.Series([0.5, 0.5])
elif direction == 'increasing':
if c < a:
return pd.Series([0, 1])
elif a < c:
return pd.Series([1, 0])
else:
return pd.Series([0.5, 0.5])
else:
return pd.Series([None, None]) # In case metric not found
df_both[['winner_Component', 'winner_Atlas']] = df_both.apply(get_winners, axis=1)
def get_nodes_edges(row):
trj_name = row['booted_trajectory']
len_nodes, len_edges = trj_size_dict[trj_name]
return pd.Series([len_nodes, len_edges])
df_both[['len_nodes', 'len_edges']] = df_both.apply(get_nodes_edges, axis=1)
label_mapping_o = {
"adjacency": "Adjacency",
"embedding": "Embedding",
"pseudotime": "Pseudotime",
}
df_both["path"] = [label_mapping_o[i] for i in df_both["path"]]
# long_both
id_cols = [
"decompose_method", "trajectory_class", "trajectory",
"booted_trajectory", 'len_nodes', 'len_edges', "path", "metric",
"trajectory_annotation_column"
]
_df = df_both.reset_index()
id_cols = ["index"] + id_cols
long_both = (
pd.wide_to_long(
_df, # the original wide dataframe
stubnames=["score", "winner"],
i=id_cols, # columns that make a unique row
j="focus", # new column: 'component' or 'atlas'
sep="_", # underscore separates stub & suffix
suffix="(Component|Atlas)" # the two suffix values to capture
)
.reset_index()
.sort_values(id_cols + ["focus"]) # optional tidy ordering
)
del long_both["index"]
replacements = {
'Pseudotime': 'Cell Order\ncontinuity',
'Embedding': 'Trajectory\ndominance',
'Adjacency': 'Topology\nconservation'
}
long_both_p = long_both.pivot(index=["booted_trajectory", "path", "metric"], columns='focus', values='winner')
long_both_p = long_both_p.groupby(level=[0, 1]).mean()
long_both_p.rename(index=replacements, level="path", inplace=True)
long_both_back = (
long_both_p
.stack(dropna=False) # <- dropna=True by default
.rename('winner')
.reset_index() # gives: booted_trajectory, path, metric, focus, winner
)
long_both_back.head()
# %%
label_mapping = {
"Atlas": "IA",
"Component": "Garcia",
}
# match boxplot’s height & per-column width
fig_height = 3
n_cols = long_both_back["focus"].nunique()
col_width = 2.4
fig_width = n_cols * col_width
fig, ax = plt.subplots(figsize=(fig_width, fig_height))
# draw bars
bp = sns.barplot(
data=long_both_back,
hue="focus",
y="winner",
x="path",
palette=sns.color_palette("Greys", n_colors=len(label_mapping)+2),
width=0.8,
saturation=1,
errorbar="sd",
capsize=0.05,
err_kws={'color': 'black', 'linewidth': 0.81},
edgecolor="black",
linewidth=0.81,
zorder=1,
ax=ax
)
# enforce edge styling
for patch in bp.patches:
patch.set_edgecolor("black")
patch.set_linewidth(0.81)
# axes
ax.set_xticklabels(ax.get_xticklabels(), rotation=90, ha='center', fontsize=10)
ax.tick_params(axis="y", labelsize=10)
ax.set_xlabel('Germline', fontsize=10, labelpad=10)
plt.ylabel("Performance", fontsize=10, labelpad=10)
ax.set_ylim(0, None)
ax.yaxis.set_major_formatter(FormatStrFormatter('%.1f'))
# grid + despine
ax.grid(axis='y', linestyle='--', alpha=0.4)
sns.despine(left=True, trim=False)
# remove the old legend
ax.legend_.remove()
# build new, professional legend
handles, labels = ax.get_legend_handles_labels()
new_labels = [label_mapping.get(l, l) for l in labels]
slim_handles = [
mpatches.Patch(
facecolor=h.get_facecolor(), # <-- keep the whole RGBA tuple
edgecolor="black",
linewidth=0.5 # thinner border
)
for h in handles
]
leg = ax.legend(
slim_handles,
new_labels,
title="Representation",
loc="center left",
bbox_to_anchor=(1.02, 0.5),
alignment="left",
frameon=False,
title_fontsize=8,
fontsize=8,
borderaxespad=0.5,
handletextpad=0.4,
labelspacing=0.5,
)
# tighten layout (accounts for legend)
plt.tight_layout(rect=[0, 0, 0.85, 1])
plt.savefig("/Users/kemalinecik/Downloads/fig2fig/comparison_germline_component.pdf")
# %%
long_both_germ = long_both_back.copy()
# %%
focus_trajectory = "Haem_ref"
df_compo = result_df_main[result_df_main["trajectory"] == focus_trajectory]
df_atlas = result_df_boot[result_df_boot["trajectory"] == focus_trajectory]
df_atlas.reset_index(inplace=True, drop=True)
df_compo.reset_index(inplace=True, drop=True)
decompose_method = "2"
input_trajectories_path_litc = os.path.join(temp_move, f"adata_hdca_input_{focus_trajectory}_litc_{decompose_method}.pkl")
with open(input_trajectories_path_litc, 'rb') as _file:
lit = pickle.load(_file)
trj_size_dict = dict()
for trj_name in lit.graph['trajectories']:
trj_size_dict[trj_name] = len(lit.get_trajectory(trj_name, False).nodes()), len(lit.get_trajectory(trj_name, False).edges())
assert df_compo.shape == df_atlas.shape
for i in range(len(df_compo)):
i1 = list(df_atlas.iloc[i][["booted_trajectory", "metric"]])
i2 = list(df_compo.iloc[i][["booted_trajectory", "metric"]])
assert i1==i2
# df_both
df_both = df_compo.copy()
del df_both["score"]
df_both["score_Component"] = list(df_compo["score"])
df_both["score_Atlas"] = list(df_atlas["score"])
df_both.reset_index(inplace=True, drop=True)
del df_both["anndata_subset"]
def get_winners(row):
metric = row['metric']
direction = metric_direction_dict[metric]
c = row['score_Component']
a = row['score_Atlas']
if direction == 'decreasing':
if c > a:
return pd.Series([0, 1])
elif a > c:
return pd.Series([1, 0])
else:
return pd.Series([0.5, 0.5])
elif direction == 'increasing':
if c < a:
return pd.Series([0, 1])
elif a < c:
return pd.Series([1, 0])
else:
return pd.Series([0.5, 0.5])
else:
return pd.Series([None, None]) # In case metric not found
df_both[['winner_Component', 'winner_Atlas']] = df_both.apply(get_winners, axis=1)
def get_nodes_edges(row):
trj_name = row['booted_trajectory']
len_nodes, len_edges = trj_size_dict[trj_name]
return pd.Series([len_nodes, len_edges])
df_both[['len_nodes', 'len_edges']] = df_both.apply(get_nodes_edges, axis=1)
label_mapping_o = {
"adjacency": "Adjacency",
"embedding": "Embedding",
"pseudotime": "Pseudotime",
}
df_both["path"] = [label_mapping_o[i] for i in df_both["path"]]
# long_both
id_cols = [
"decompose_method", "trajectory_class", "trajectory",
"booted_trajectory", 'len_nodes', 'len_edges', "path", "metric",
"trajectory_annotation_column"
]
_df = df_both.reset_index()
id_cols = ["index"] + id_cols
long_both = (
pd.wide_to_long(
_df, # the original wide dataframe
stubnames=["score", "winner"],
i=id_cols, # columns that make a unique row
j="focus", # new column: 'component' or 'atlas'
sep="_", # underscore separates stub & suffix
suffix="(Component|Atlas)" # the two suffix values to capture
)
.reset_index()
.sort_values(id_cols + ["focus"]) # optional tidy ordering
)
del long_both["index"]
replacements = {
'Pseudotime': 'Cell Order\ncontinuity',
'Embedding': 'Trajectory\ndominance',
'Adjacency': 'Topology\nconservation'
}
long_both_p = long_both.pivot(index=["booted_trajectory", "path", "metric"], columns='focus', values='winner')
long_both_p = long_both_p.groupby(level=[0, 1]).mean()
long_both_p.rename(index=replacements, level="path", inplace=True)
long_both_back = (
long_both_p
.stack(dropna=False) # <- dropna=True by default
.rename('winner')
.reset_index() # gives: booted_trajectory, path, metric, focus, winner
)
long_both_back.head()
# %%
label_mapping = {
"Atlas": "IA",
"Component": "Suo",
}
# match boxplot’s height & per-column width
fig_height = 3
n_cols = long_both_back["focus"].nunique()
col_width = 2.4
fig_width = n_cols * col_width
fig, ax = plt.subplots(figsize=(fig_width, fig_height))
# draw bars
bp = sns.barplot(
data=long_both_back,
hue="focus",
y="winner",
x="path",
palette=sns.color_palette("Greys", n_colors=len(label_mapping)+2),
width=0.8,
saturation=1,
errorbar="sd",
capsize=0.05,
err_kws={'color': 'black', 'linewidth': 0.81},
edgecolor="black",
linewidth=0.81,
zorder=1,
ax=ax
)
# enforce edge styling
for patch in bp.patches:
patch.set_edgecolor("black")
patch.set_linewidth(0.81)
# axes
ax.set_xticklabels(ax.get_xticklabels(), rotation=90, ha='center', fontsize=10)
ax.tick_params(axis="y", labelsize=10)
ax.set_xlabel('Hematopoietic Lineage', fontsize=10, labelpad=10)
plt.ylabel("Performance", fontsize=10, labelpad=10)
ax.set_ylim(0, None)
ax.yaxis.set_major_formatter(FormatStrFormatter('%.1f'))
# grid + despine
ax.grid(axis='y', linestyle='--', alpha=0.4)
sns.despine(left=True, trim=False)
# remove the old legend
ax.legend_.remove()
# build new, professional legend
handles, labels = ax.get_legend_handles_labels()
new_labels = [label_mapping.get(l, l) for l in labels]
slim_handles = [
mpatches.Patch(
facecolor=h.get_facecolor(), # <-- keep the whole RGBA tuple
edgecolor="black",
linewidth=0.5 # thinner border
)
for h in handles
]
leg = ax.legend(
slim_handles,
new_labels,
title="Representation",
loc="center left",
bbox_to_anchor=(1.02, 0.5),
alignment="left",
frameon=False,
title_fontsize=8,
fontsize=8,
borderaxespad=0.5,
handletextpad=0.4,
labelspacing=0.5,
)
# tighten layout (accounts for legend)
plt.tight_layout(rect=[0, 0, 0.85, 1])
plt.savefig("/Users/kemalinecik/Downloads/fig2fig/comparison_hemato_component.pdf")
# %%
long_both_haem = long_both_back.copy()
# %%
long_both_haem["trajectory_class"] = "Hematopoietic\nLineage"
long_both_germ["trajectory_class"] = "Germline"
long_both2 = pd.concat([long_both_haem, long_both_germ])
long_both2.reset_index(drop=True, inplace=True)
# %%
label_mapping = {
"Atlas": "Atlas",
"Component": "Component",
}
# match boxplot’s height & per-column width
fig_height = 3
n_cols = long_both2["focus"].nunique()
col_width = 2
fig_width = n_cols * col_width
fig, ax = plt.subplots(figsize=(fig_width, fig_height))
# draw bars
bp = sns.barplot(
data=long_both2,
hue="focus",
y="winner",
x="trajectory_class",
palette=sns.color_palette("Greys", n_colors=len(label_mapping)+2),
width=0.8,
saturation=1,
errorbar="sd",
capsize=0.05,
err_kws={'color': 'black', 'linewidth': 0.81},
edgecolor="black",
linewidth=0.81,
zorder=1,
ax=ax
)
# enforce edge styling
for patch in bp.patches:
patch.set_edgecolor("black")
patch.set_linewidth(0.81)
# axes
ax.set_xticklabels(ax.get_xticklabels(), rotation=90, ha='center', fontsize=10)
ax.tick_params(axis="y", labelsize=10)
ax.set_xlabel('Lineage', fontsize=10, labelpad=10)
plt.ylabel("Performance", fontsize=10, labelpad=10)
ax.set_ylim(0, None)
ax.yaxis.set_major_formatter(FormatStrFormatter('%.1f'))
# grid + despine
ax.grid(axis='y', linestyle='--', alpha=0.4)
sns.despine(left=True, trim=False)
# remove the old legend
ax.legend_.remove()
# build new, professional legend
handles, labels = ax.get_legend_handles_labels()
new_labels = [label_mapping.get(l, l) for l in labels]
slim_handles = [
mpatches.Patch(
facecolor=h.get_facecolor(), # <-- keep the whole RGBA tuple
edgecolor="black",
linewidth=0.5 # thinner border
)
for h in handles
]
leg = ax.legend(
slim_handles,
new_labels,
title="Representation",
loc="center left",
bbox_to_anchor=(1.02, 0.5),
alignment="left",
frameon=False,
title_fontsize=8,
fontsize=8,
borderaxespad=0.5,
handletextpad=0.4,
labelspacing=0.5,
)
# tighten layout (accounts for legend)
plt.tight_layout(rect=[0, 0, 0.85, 1])
plt.savefig("/Users/kemalinecik/Downloads/fig2fig/comparison_germline_hemato_both.pdf")
# %%
1
# %% [markdown]
# # Figure UMAPs
# %%
import scvelo as scv
import cellrank as cr
import scanpy.external as sce
scv.settings.verbosity = 3
cr.settings.verbosity = 2
sc.settings.verbose = 3
# %%
adata_comp = ad.read_h5ad(os.path.join(temp_move, f"anndata_hdca_input_germline_litc_comparison_comp.h5ad"))
adata_atlas = ad.read_h5ad(os.path.join(temp_move, f"anndata_hdca_input_germline_litc_comparison_atlas.h5ad"))
# %%
display((adata_atlas.shape, adata_comp.shape))
sc.pp.neighbors(adata_atlas, n_neighbors=15)
sc.tl.umap(adata_atlas)
sc.pp.neighbors(adata_comp, n_neighbors=15)
sc.tl.umap(adata_comp)
# %%
(
df_hdca_component_adata_paths,
df_hdca_lvl3_adata_paths,
out_df,
ground_truth_trajectories,
ground_truth_trajectories_annotation_level,
trajectory_subset_dict
) = load_pickle_bundle(os.path.join(temp_move, "_experiment_development_atlas_helper_pickle_bundle.pkl"))
# %%
# Mapping from original to formal names
celltype_mapping = {
"GC": "Granulosa Cell",
"PGC": "Primordial Germ Cell",
"oocyte": "Oocyte",
"oogonia": "Oogonia",
"oogonia_meiotic_1": "Oogonia\n(Meiotic I)",
"oogonia_meiotic_2": "Oogonia\n(Meiotic II)",
"pre_oocyte": "Pre-Oocyte",
"pre_spermatogonia": "Pre-Spermatogonia"
}
# Apply mapping to the annotations
adata_atlas.obs[ground_truth_trajectories_annotation_level[focus_trajectory]] = \
adata_atlas.obs[ground_truth_trajectories_annotation_level[focus_trajectory]].map(celltype_mapping)
# Apply mapping to the annotations
adata_comp.obs[ground_truth_trajectories_annotation_level[focus_trajectory]] = \
adata_comp.obs[ground_truth_trajectories_annotation_level[focus_trajectory]].map(celltype_mapping)
# %%
def cr_calc(adata_cr, root_cell_type, i, ann_levl, knn=15):
sc.tl.pca(adata_cr, n_comps=11, svd_solver='arpack')
sce.tl.palantir(adata_cr, n_components=10, knn=knn)
root_cells = adata_cr.obs_names[adata_cr.obs[ann_levl] == root_cell_type]
if len(root_cells) == 0:
raise ValueError(f"No cells found with {ann_levl!r} == {root_cell_type!r}")
start_cell = root_cells[i]
pr_res = sce.tl.palantir_results(
adata_cr,
early_cell=start_cell,
ms_data="X_palantir_multiscale",
knn=knn,
num_waypoints=1200,
n_jobs=-1
)
adata_cr.obs["palantir_pseudotime"] = pr_res.pseudotime
pk = cr.kernels.PseudotimeKernel(adata_cr, time_key='palantir_pseudotime')
pk.compute_transition_matrix()
return pk
# %%
focus_trajectory = "germline"
ann_lvl = ground_truth_trajectories_annotation_level[focus_trajectory]
pk_adata_atlas = cr_calc(adata_atlas.copy(), root_cell_type = "Primordial Germ Cell", i=10, ann_levl=ann_lvl)
# %%
plt.figure(figsize=(4,3))
pk_adata_atlas.plot_projection(basis='umap', recompute=False, color=[ann_lvl], title="", ax=plt.gca(), legend_fontsize="medium", legend_fontweight="bold", legend_fontoutline=2,
save="/Users/kemalinecik/Downloads/fig2fig/comparison_germline_atlas_umap.pdf")
# %%
pk_adata_comp = cr_calc(adata_comp.copy(), root_cell_type = "Primordial Germ Cell", i=10, ann_levl=ann_lvl)
# %%
plt.figure(figsize=(4,3))
pk_adata_comp.plot_projection(basis='umap', recompute=False, color=[ann_lvl], title="", ax=plt.gca(), legend_fontsize="medium", legend_fontweight="bold", legend_fontoutline=2,
save="/Users/kemalinecik/Downloads/fig2fig/comparison_germline_component_umap.pdf")
# %% [markdown]
# # Figure UMAPs
# %%
import scvelo as scv
import cellrank as cr
import scanpy.external as sce
scv.settings.verbosity = 3
cr.settings.verbosity = 2
sc.settings.verbose = 3
(
df_hdca_component_adata_paths,
df_hdca_lvl3_adata_paths,
out_df,
ground_truth_trajectories,
ground_truth_trajectories_annotation_level,
trajectory_subset_dict
) = load_pickle_bundle(os.path.join(temp_move, "_experiment_development_atlas_helper_pickle_bundle.pkl"))
# %%
def cr_calc(adata_cr, root_cell_type, i, ann_levl, knn=15):
sc.tl.pca(adata_cr, n_comps=11, svd_solver='arpack')
sce.tl.palantir(adata_cr, n_components=10, knn=knn)
root_cells = adata_cr.obs_names[adata_cr.obs[ann_levl] == root_cell_type]
if len(root_cells) == 0:
raise ValueError(f"No cells found with {ann_levl!r} == {root_cell_type!r}")
start_cell = root_cells[i]
pr_res = sce.tl.palantir_results(
adata_cr,
early_cell=start_cell,
ms_data="X_palantir_multiscale",
knn=knn,
num_waypoints=1200,
n_jobs=-1
)
adata_cr.obs["palantir_pseudotime"] = pr_res.pseudotime
pk = cr.kernels.PseudotimeKernel(adata_cr, time_key='palantir_pseudotime')
pk.compute_transition_matrix()
return pk
# %%
# Mapping for B-cell and progenitor annotations
bcell_mapping = {
"Mature_B": "Mature B",
"Large_pre_B": "Large Pre B",
"Small_pre_B": "Small Pre B",
"B1": "B1",
"Pro_B": "Pro B",
"Late_Pro_B": "Late Pro B",
"HSC_MPP": "HSC MPP",
"Immature_B": "Immature B",
"Pre_pro_B": "Pre-pro B",
"LMPP_MLP": "LMPP MLP",
"CYCLING_B": "Cycling B",
"Plasma_B": "Plasma B"
}
# %%
for use_rep in ['scpoli', 'scvi']:
print("\n\n\n\%%%%%%%%%%%%%%\n\n\n", use_rep)
adata_atlas = ad.read_h5ad(os.path.join(temp_move, f"anndata_hdca_input_Haem_ref_litc_2_{use_rep}_nodes_b.h5ad"))
print(adata_atlas.shape)
sc.pp.subsample(adata_atlas, fraction=0.2)
print(adata_atlas.shape)
sc.pp.neighbors(adata_atlas, n_neighbors=15)
sc.tl.umap(adata_atlas)
ann_lvl = ground_truth_trajectories_annotation_level["Haem_ref"]
adata_atlas.obs[ann_lvl] = adata_atlas.obs[ann_lvl].map(bcell_mapping)
pk_adata_atlas = cr_calc(adata_atlas.copy(), root_cell_type = "HSC MPP", i=10, ann_levl=ann_lvl, knn=30)
plt.figure(figsize=(5,4))
pk_adata_atlas.plot_projection(basis='umap', recompute=False, color=[ann_lvl], title="", ax=plt.gca(), legend_fontsize="medium", legend_fontweight="bold", legend_fontoutline=2,
save=f"/Users/kemalinecik/Downloads/fig2fig/comparison_models_b_umap_{use_rep}.svg")
# break
del adata_atlas, pk_adata_atlas
gc.collect()
# %%

sctram_figures_only.ipynb at commit e570de5, under BSD-3-Clause · at the source

Overview

Authors: Simone Webb^1,2, Antony Rose^1,2, Chuan Xu³, Lloyd Steele², Masami A Kuri², Emily Stephenson^1,2, Kemal Inecik⁴, Daniyal Jafree^2,5, April R Foster², Daniela Basurto-Lozada^1,2, Nana-Jane Chipampe², Anna V Pournara², Marc-Antoine Jacques^6,7, Ken To^2,3,8, Chloe Admane^1,2, Efpraxia Kritikaki^1,2, Marta A Chroscik^1,2, David Horsfall^1,2, Julia Foreman⁶, Koen Rademaker²

and 74 other authors

Julia Karjalainen⁹, Anna Laddach¹⁰, Shaista Madad², John EG Lawrence^2,8, Vitalii Kleshchevnikov², Steven Lisgo¹, Jimmy Tsz Hang Lee², Joséphine Blevinal¹¹, Ahlam Alqahtani¹, Stanislaw Makarchuk², Jake Jackson¹, Ekin Ucuncu⁵, Teresa P Silva⁵, Valentina Lorenzi^2,6, Fereshteh Torabi², Rachel A Botting¹, Kenny Roberts², Bayanne Olabi¹, Keerthi P Chakala², Leander Dony^4,12, Gabriele Dall’Aglio⁴, Ana-Maria Cujba³, Holly J Whitfield², Christine Hale², Justin Engelbert¹, Tong Li², Elena Prigmore², Nusayhah Hudaa Gopee^1,2, Michael W Mather¹, Ben Talks¹, Lijiang Fei³, Diana Pereira¹³, Corina Moldovan¹⁴, Rasa Elmentaite³, Krzysztof Polanski³, Alexander V Predeus², Pasha Mazin², Vicky Rowe², Semih Bayraktar³, Vincent R Knight-Schrijver³, Menna Clatworthy¹⁵, Kazumasa Kanemaru³, James Cranley^3,16, Andrew Guo¹⁵, Sanjay Sinha³, David McDonald¹, Andrew Filby¹, Janet Kerwin¹, Peng He^17,18, Inês Sequeira¹³, Roser Vento-Tormo², Malte D. Luecken^4,19, Alain Chédotal^11,20,21, Deanne Taylor^22,23, Silvia Martin-Almedina²⁴, Pia Ostergaard²⁴, Sam Behjati^2,25, Elizabeth Robertson²⁶, Nicola K Wilson^3,7, John Marioni^5,27, Helen V Firth^2,28, Caroline F Wright²⁹, Olivier Pourquie³⁰, Laure Gambardella², Paula Alexandre⁵, Berthold Gottgens^3,7, Kerstin B Meyer², MS Vijayabaskar², Mo Lotfollahi², Sarah A Teichmann^3,16,31, Fabian J Theis⁴, Fränze Progatzky⁹, Omer Ali Bayraktar², Muzlifah Haniffa^2,28,32

ORCID iDs: Simone Webb, Antony Rose, Emily Stephenson, Kemal Inecik, Daniyal Jafree, Daniela Basurto-Lozada, Teresa P Silva, Laure Gambardella, Paula Alexandre, Fränze Progatzky

32 affiliations

Biosciences Institute, Newcastle University, Newcastle upon Tyne, UK
Wellcome Sanger Institute, Wellcome Genome Campus, Hinxton, UK
Cambridge Stem Cell Institute, Jeffrey Cheah Biomedical Centre, Cambridge Biomedical Campus, University of Cambridge, Cambridge, UK
Institute of Computational Biology, Computational Health Center, Helmholtz Munich, Neuherberg, Germany
Developmental Biology & Cancer Department, University College London (UCL) Great Ormond Street Institute of Child Health, UCL, London, UK
European Molecular Biology Laboratory, European Bioinformatics Institute (EMBL-EBI), Wellcome Genome Campus, Cambridge, UK
Department of Haematology, University of Cambridge, Cambridge, UK
Department of Surgery, University of Cambridge, Cambridge, UK
Kennedy Institute of Rheumatology, University of Oxford, Oxford, UK
Nervous System Development and Homeostasis Laboratory, the Francis Crick Institute, London, UK
Sorbonne Université, INSERM, CNRS, Institut de la Vision, Paris, France
Department of Biosystems Science and Engineering, ETH Zürich, Basel, Switzerland
Center for Oral Immunobiology and Regenerative Medicine, Barts Centre for Squamous Cancer, Institute of Dentistry, Queen Mary University of London, London, UK
Department of Pathology, Newcastle Hospitals NHS Foundation Trust, Newcastle upon Tyne, UK
Molecular Immunity Unit, Department of Medicine, University of Cambridge, Cambridge, UK
Department of Medicine, University of Cambridge, Cambridge, UK
Department of Pathology, University of California, San Francisco, CA, USA
Eli and Edythe Broad Center of Regeneration Medicine and Stem Cell Research, University of California, San Francisco, San Francisco, CA, USA
Institute of Lung Health and Immunity (LHI), Helmholtz Munich, Comprehensive Pneumology Center (CPC-M), Germany; Member of the German Center for Lung Research (DZL)
Institut de Pathologie, Groupe Hospitalier Est, Hospices Civils de Lyon, Lyon, France
University Claude Bernard Lyon 1, MeLiS, CNRS UMR5284, INSERM U1314, 69008, Lyon, France
Department of Biomedical and Health Informatics (DBHi), The Children’s Hospital of Philadelphia, Philadelphia, PA, USA
Department of Pediatrics, University of Pennsylvania Perelman School of Medicine, Philadelphia, PA, USA
School of Health and Medical Sciences, City St George’s University of London, London, UK
Department of Paediatrics, University of Cambridge, Cambridge, UK
Sir William Dunn School of Pathology, University of Oxford, Oxford, UK
Cancer Research UK, Cambridge Institute, University of Cambridge, Cambridge, UK
Cambridge University Hospitals Foundation Trust, Addenbrooke’s Hospital, Cambridge, UK
Department of Clinical and Biomedical Sciences, St Luke’s Campus, University of Exeter, Exeter, UK
Department of Genetics, Harvard Medical School and Department of Pathology, Brigham and Women’s Hospital, Boston, MA, USA
CIFAR Macmillan Multi-scale Human Programme, CIFAR, Toronto, Canada
Cambridge Institute for Therapeutic Immunology and Infectious Diseases (CIITID), University of Cambridge, UK

Institutions: Newcastle University (United Kingdom); Wellcome Sanger Institute (United Kingdom); University of Cambridge (United Kingdom); Helmholtz Munich (Germany); Institute of Computational Biology (Germany); University College London (United Kingdom); European Bioinformatics Institute (United Kingdom); Kennedy Institute of Rheumatology (United Kingdom); University of Oxford (United Kingdom); The Francis Crick Institute (United Kingdom); Sorbonne Université (France); Inserm (France); Centre National de la Recherche Scientifique (France); Institut de la Vision (France); ETH Zurich (Switzerland); Queen Mary University of London (United Kingdom); Newcastle upon Tyne Hospitals NHS Foundation Trust (United Kingdom); University of California, San Francisco (United States); German Center for Lung Research (Germany); Hospices Civils de Lyon (France); Children's Hospital of Philadelphia (United States); University of Pennsylvania (United States); City St George's, University of London (United Kingdom); Cancer Research UK (United Kingdom); Addenbrooke's Hospital (United Kingdom); University of Exeter (United Kingdom); Brigham and Women's Hospital (United States)

Journal: bioRxiv (preprint)

Dates: published online 1 April 2026

Type: Preprint

License: CC BY

Identifiers: DOI 10.64898/2026.03.30.714220 · OpenAlex W7147094455

Open access: green, a free copy (OpenAlex)

Status: code verified

Categories: human (organism)

Methods: Preprocessing, Connectivity, Statistics, Smoothing, state filtering, decompositions, Machine learning, fMRI & imaging

Topic: Single-cell and spatial transcriptomics (Molecular Biology, Biochemistry, Genetics and Molecular Biology), according to OpenAlex

Funding: National Institute for Health Research (NIHR) (NIHR203312); Wellcome Trust (WT223718/Z/21/Z, 203151/Z/16/Z, 226083/Z/22/Z, 203151/A/16/Z, 314710/Z/24/Z); British Heart Foundation (RG/17/7/33217, SP/13/5/30288); Biotechnology and Biological Sciences Research Council (BB/Y003179/1); Versus Arthritis (22981)

Citations: cited by 2 papers (Europe PMC); 193 references in the paper

Abstract

Single cell genomics has enabled analysis of human prenatal development at unprecedented resolution. However, most studies have relied on dissociated tissues during restricted windows of development, limiting insights into how spatially distributed networks of cells, and multicellular niches emerge and adapt to distinct organ microenvironments in situ. Moreover, existing human developmental atlases have not yet been harmonised, and we thus lack a comprehensive catalogue of known cell types in the developing human body.

Here, we introduce the Human Developmental Cell Atlas (HDCA), a unified structural, cellular and molecular resource for prenatal human development. The HDCA integrates published and unpublished single cell/nucleus RNAseq atlases across prenatal organs, and includes a newly generated, spatially resolved, multimodal cell atlas of intact human embryos. Spanning 4-22 post conceptional weeks, capturing embryonic and early to mid fetal stages, the HDCA contains ~4.6 million cells/nuclei which resolve into ~450 cell types, explorable with a bespoke web portal.

For a global overview of the human embryo’s multicellular communities, we applied unsupervised deep learning to our intact human embryo spatial data, charting 114 tissue niches that are structural and signalling hubs for the cellular interactions of the embryo. Guided by these niches, we profiled cellular networks over space and time, not examinable using single-organ atlases. In so doing, we revealed tissue-specific fibroblast patterning from previously undescribed mesenchyme progenitors, early diversification of organ-specific blood capillaries and lymphatic vasculature, emergence of neural crest cell fates, the formation of placode- and neural crest-derived peripheral sensory neurons, and how tissue niches guide peripheral neuron maturation and axonal migration. The HDCA thus serves as a comprehensive step towards a comprehensive understanding of human prenatal development, and a template towards unravelling the biology of congenital disorders.

Reproduced under the paper's license (CC BY), from the paper cited above.

Repositories

Its files are read in the Code ↔ Paper reader above, with 19 matches between paragraphs and lines of code.

haniffalab/hdca-paper

License: none: the authors keep all their rights

State: the link is dead, verified on 28 September 2026

Evidence: found in the paper

Software Heritage: not archived

Found in: “Code”

Not found: README, license file, CITATION.cff, environment file, tests, continuous integration, documentation

Availability: 1 check, the latest on 28 September 2026: the link is dead

28 September 2026: the link is dead

At the source: github.com/haniffalab/hdca-paper

theislab/idtrack

License: BSD-3-Clause

State: the link answers, verified on 28 September 2026

Evidence: files inventoried

Commit: d8514ab51a494842f4278e769c9538f103592c7a, 20 January 2026

Languages: Python (72), Jupyter (38), Shell (8)

Size: 180 files, 118 scripts

Software Heritage: archived

Found in: “Code”

Holds: README, license file, environment (poetry.lock, pyproject.toml, docs/requirements.txt), tests, continuous integration, documentation, 38 notebooks

Not found: CITATION.cff

Tools: pandas (40 files), NumPy (31 files), Matplotlib (20 files), seaborn (18 files), anndata (7 files), NetworkX (5 files), SciPy (4 files), Scanpy (3 files), h5py (2 files), Biopython (1 file), PyTorch (1 file), Hugging Face Transformers (1 file)

Availability: 1 check, the latest on 28 September 2026: the link answers

28 September 2026: the link answers

63 files

docs/_notebooks/00_idtrack_overview.ipynb, Jupyter, 264 lines
docs/_notebooks/01_installation_guide.ipynb, Jupyter, 187 lines
docs/_notebooks/02_prepare_new_external_yaml.ipynb, Jupyter, 627 lines
docs/_notebooks/03_initialization_graph.ipynb, Jupyter, 289 lines
docs/_notebooks/04_api_deep_dive_human.ipynb, Jupyter, 307 lines
docs/_notebooks/05_tutorial_harmonization.ipynb, Jupyter, 575 lines
docs/_notebooks/06_tutorial_humanization_mouse_pig_to_human.ipynb, Jupyter, 240 lines
docs/_notebooks/07_advanced_topics.ipynb, Jupyter, 431 lines
docs/_notebooks/99_initialization_test.ipynb, Jupyter, 185 lines
docs/_notebooks/_notebook_utils.py, Python, 381 lines
docs/conf.py, Python, 215 lines
example_manual_running copy.ipynb, Jupyter, 194 lines
idtrack/__init__.py, Python, 51 lines
idtrack/__main__.py, Python, 125 lines
idtrack/_api.py, Python, 697 lines
idtrack/_connection_bridge.py, Python, 431 lines
idtrack/_database_manager.py, Python, 3,045 lines
idtrack/_db.py, Python, 420 lines
idtrack/_external_databases.py, Python, 385 lines
idtrack/_external_mappers/__init__.py, Python, 64 lines
idtrack/_external_mappers/_backend_gget.py, Python, 270 lines
idtrack/_external_mappers/_backend_gprofiler.py, Python, 398 lines
idtrack/_external_mappers/_backend_mygene.py, Python, 277 lines
idtrack/_external_mappers/_backend_pybiomart.py, Python, 519 lines
idtrack/_external_mappers/_constants.py, Python, 283 lines
idtrack/_external_mappers/_convert.py, Python, 157 lines
idtrack/_external_mappers/_ortholog.py, Python, 663 lines
idtrack/_external_mappers/_utils.py, Python, 534 lines
idtrack/_graph_maker.py, Python, 1,160 lines
idtrack/_harmonize_features.py, Python, 1,175 lines, 4 matches
idtrack/_the_graph.py, Python, 1,369 lines
idtrack/_track.py, Python, 2,300 lines
idtrack/_track_tests.py, Python, 1,174 lines
idtrack/_utils_hdf5.py, Python, 772 lines
idtrack/_verify_organism.py, Python, 200 lines
noxfile.py, Python, 329 lines
reproducibility/experiments/example_manual_running.ipynb, Jupyter, 342 lines
reproducibility/experiments/experiment_cellranger_idtrack/Untitled.ipynb, Jupyter, 10 lines
reproducibility/experiments/experiment_cellranger_idtrack/analysis copy.ipynb, Jupyter, 3,325 lines
reproducibility/experiments/experiment_cellranger_idtrack/analysis.ipynb, Jupyter, 3,158 lines
reproducibility/experiments/experiment_cellranger_idtrack/analysis_gold_standard_expression_consistency.ipynb, Jupyter, 557 lines
reproducibility/experiments/experiment_cellranger_idtrack/analysis_gold_standard_fig4c.ipynb, Jupyter, 797 lines
reproducibility/experiments/experiment_cellranger_idtrack/create_data.ipynb, Jupyter, 405 lines
reproducibility/experiments/experiment_external_bridges/00_external_bridges_tradeoff.ipynb, Jupyter, 376 lines
reproducibility/experiments/experiment_hlca/00_hlca_manuscript_table1_and_figures.ipynb, Jupyter, 1,580 lines
reproducibility/experiments/experiment_identifier_drift/00_identifier_drift_case_studies.ipynb, Jupyter, 506 lines
reproducibility/experiments/experiment_marketing_overview/00_marketing_overview_dashboard.ipynb, Jupyter, 194 lines
reproducibility/experiments/experiment_other_organisms/00_other_organisms_showcase.ipynb, Jupyter, 279 lines
reproducibility/experiments/experiment_other_organisms/01_other_organisms_deep_dive_time_axis.ipynb, Jupyter, 351 lines
reproducibility/experiments/experiment_random_stress_tests/00_random_stress_tests_fig4ab.ipynb, Jupyter, 759 lines
reproducibility/experiments/experiment_time_travel_matrix/00_build_time_travel_matrix_cache.ipynb, Jupyter, 291 lines
reproducibility/experiments/experiment_time_travel_matrix/01_analyze_time_travel_matrix.ipynb, Jupyter, 654 lines
reproducibility/experiments/experiment_time_travel_matrix/02_roundtrip_consistency.ipynb, Jupyter, 424 lines
reproducibility/experiments/experiment_time_travel_vs_external_mappers/00_build_time_travel_vs_external_mappers_cache.ipynb, Jupyter, 466 lines
reproducibility/experiments/experiment_time_travel_vs_external_mappers/01_analyze_time_travel_vs_external_mappers.ipynb, Jupyter, 478 lines
reproducibility/experiments/experiment_time_travel_vs_external_mappers/02_case_studies_external_mapper_failures.ipynb, Jupyter, 388 lines
reproducibility/experiments/experiment_tool_comparison/00_tool_comparison_capability_matrix_fig4d.ipynb, Jupyter, 561 lines
reproducibility/experiments/experiment_tool_comparison/01_external_mapper_variability_deep_dive.ipynb, Jupyter, 606 lines
reproducibility/experiments/experiment_tool_comparison/02_pybiomart_release_sensitivity.ipynb, Jupyter, 375 lines
reproducibility/experiments/figure_rcparams/rcparams.ipynb, Jupyter, 26 lines
reproducibility/experiments/prev/example_manual_running.ipynb, Jupyter, 287 lines
repository limit reached (2,000 files or 30 MB): the rest is at the source (57 files)
LICENSE, License, 29 lines
README.rst, Text, 80 lines

At the source: github.com/theislab/idtrack

cellgeni/OMERO_tools

License: none: the authors keep all their rights

State: the link answers, verified on 28 September 2026

Evidence: files inventoried

Commit: f4d9d3f1633265743936db2cfe48b53eda992272, 28 July 2026

Languages: Python (8)

Size: 14 files, 8 scripts

Software Heritage: not archived

Found in: the text, “Whole embryo: manual annotation of H&E images, a”

Holds: README, environment (environment.yml)

Not found: license file, CITATION.cff, tests, continuous integration, documentation

Tools: NumPy (6 files), pandas (6 files), Pillow (6 files), Scanpy (5 files), Squidpy (5 files), Matplotlib (4 files)

Availability: 1 check, the latest on 28 September 2026: the link answers

28 September 2026: the link answers

9 files

ann2SR.py, Python, 472 lines
ann2Xenium.py, Python, 237 lines
ann2Xenium_batch.py, Python, 49 lines
ann2Xenium_coord.py, Python, 769 lines
ann2Xenium_v2.py, Python, 184 lines
categorical_ann2SR.py, Python, 193 lines
duplicate.py, Python, 291 lines
transfer_annotations_different_OMERO_servers.py, Python, 210 lines
README.md, Text, 62 lines

At the source: github.com/cellgeni/OMERO_tools

haniffalab/cherita-react

License: MIT

State: the link answers, verified on 28 September 2026

Evidence: files inventoried

Commit: 740eaf25038452adf553d2528c0c8e8bcbd62906, 8 May 2026

Languages: JavaScript (27)

Size: 150 files, 27 scripts

Software Heritage: not archived

Found in: the text, “HDCA web portal integration with developmental d”

Holds: README, license file, CITATION.cff, tests, continuous integration, documentation

Not found: environment file

Availability: 1 check, the latest on 28 September 2026: the link answers

28 September 2026: the link answers

29 files

babel.config.js, JavaScript, 18 lines
jest.config.js, JavaScript, 21 lines
prepare-release.js, JavaScript, 18 lines
sites/demo/src/components/Footer.js, JavaScript, 13 lines
sites/demo/src/index.js, JavaScript, 16 lines
sites/demo/vite.config.js, JavaScript, 40 lines
src/lib/components/scatterplot/ScatterplotLayer.js, JavaScript, 198 lines
src/lib/constants/colorscales.js, JavaScript, 138 lines
src/lib/constants/constants.js, JavaScript, 111 lines
src/lib/helpers/color-helper.js, JavaScript, 83 lines
src/lib/helpers/map-helper.js, JavaScript, 50 lines
src/lib/helpers/zarr-helper.js, JavaScript, 104 lines
src/lib/index.js, JavaScript, 32 lines
src/lib/utils/Filter.js, JavaScript, 210 lines
src/lib/utils/Resolver.js, JavaScript, 294 lines
src/lib/utils/errors.js, JavaScript, 23 lines
src/lib/utils/parquetData.js, JavaScript, 55 lines
src/lib/utils/requests.js, JavaScript, 104 lines
src/lib/utils/search.js, JavaScript, 67 lines
src/lib/utils/string.js, JavaScript, 63 lines
src/lib/utils/zarrData.js, JavaScript, 159 lines
src/lib/views/ObservationFeature/index.js, JavaScript, 7 lines
src/lib/views/PerturbationMap/index.js, JavaScript, 7 lines
src/lib/workers/scatterplotData.js, JavaScript, 45 lines
website/docusaurus.config.js, JavaScript, 116 lines
website/sidebars.js, JavaScript, 35 lines
website/src/pages/index.js, JavaScript, 5 lines
LICENSE, License, 21 lines
README.md, Text, 36 lines

At the source: github.com/haniffalab/cherita-react

theislab/sctram

License: BSD-3-Clause

State: the link answers, verified on 28 September 2026

Evidence: files inventoried

Commit: e570de5f9ec6dc783542b82c4f88b192677d4347, 10 April 2026

Languages: Python (125), Jupyter (53), Shell (6)

Size: 235 files, 184 scripts

Software Heritage: not archived

Found in: “Code”

Holds: README, license file, environment (poetry.lock, pyproject.toml, docs/requirements.txt), tests, continuous integration, documentation, 53 notebooks

Not found: CITATION.cff

Tools: anndata (18 files), Scanpy (17 files), NetworkX (16 files), NumPy (16 files), Matplotlib (15 files), pandas (15 files), seaborn (15 files), SciPy (5 files), PyTorch (4 files), scikit-learn (3 files), CuPy (1 file), scVelo (1 file), statsmodels (1 file), UMAP (1 file)

Availability: 1 check, the latest on 28 September 2026: the link answers

28 September 2026: the link answers

35 files

docs/_notebooks/_doctest.ipynb, Jupyter, 13 lines
docs/conf.py, Python, 181 lines
noxfile.py, Python, 235 lines
reproducibility/figure_rcparams/rcparams.ipynb, Jupyter, 23 lines
reproducibility/local/experiments/sctram_figures_only.ipynb, Jupyter, 1,813 lines, 6 matches
reproducibility/modules/module_api/example_operations.ipynb, Jupyter, 90 lines
reproducibility/modules/module_evaluate/example_operations_adjacency.ipynb, Jupyter, 152 lines, 1 match
reproducibility/modules/module_evaluate/example_operations_converters.ipynb, Jupyter, 233 lines
reproducibility/modules/module_evaluate/example_operations_embedding.ipynb, Jupyter, 134 lines
reproducibility/modules/module_evaluate/example_operations_pseudotimevalues.ipynb, Jupyter, 222 lines
reproducibility/modules/module_evaluate/example_operations_spatial_embedding.ipynb, Jupyter, 134 lines
reproducibility/modules/module_evaluate/example_operations_spatial_pseudotime.ipynb, Jupyter, 160 lines
reproducibility/modules/module_generate/submodule_real/example_operations.ipynb, Jupyter, 83 lines
reproducibility/modules/module_generate/submodule_synthetic_simple/example_operations.ipynb, Jupyter, 400 lines, 2 matches
reproducibility/modules/module_infer/example_operations_embedding.ipynb, Jupyter, 96 lines
reproducibility/modules/module_infer/example_operations_inference.ipynb, Jupyter, 224 lines
reproducibility/modules/module_input/bootstapping_plotting.ipynb, Jupyter, 78 lines
reproducibility/modules/module_input/example_operations.ipynb, Jupyter, 437 lines
reproducibility/modules/put_somewhere.ipynb, Jupyter, 86 lines
reproducibility/preprocessing_real_data/norman_sciplex_cpa.ipynb, Jupyter, 283 lines
reproducibility/preprocessing_real_data/other_developmental_completal.ipynb, Jupyter, 57 lines
reproducibility/preprocessing_real_data/suo_developmental_complete.ipynb, Jupyter, 112 lines
reproducibility/scripts/check_occurances_for_version.py, Python, 59 lines
reproducibility/scripts/jupyter_tunnel.sh, Shell, 68 lines
reproducibility/scripts/sync_server.sh, Shell, 270 lines
reproducibility/scripts/utils/_colors.sh, Shell, 6 lines
reproducibility/scripts/utils/_log.sh, Shell, 5 lines
reproducibility/scripts/utils/_python_environment.sh, Shell, 31 lines
reproducibility/scripts/utils/_system.py, Python, 322 lines
reproducibility/server_sync/experiments/_deprecated_suo_incremental/helper/model_training.py, Python, 101 lines, 2 matches
reproducibility/server_sync/experiments/_deprecated_suo_incremental/helper/suo_scib_iterative.py, Python, 137 lines, 1 match
reproducibility/server_sync/experiments/_deprecated_suo_incremental/helper/suo_sctram_iterative.py, Python, 89 lines
reproducibility/server_sync/experiments/_deprecated_suo_incremental/incremental_analysis.ipynb, Jupyter, 502 lines, 3 matches
repository limit reached (2,000 files or 30 MB): the rest is at the source (151 files)
LICENSE, License, 29 lines
README.rst, Text, 55 lines

At the source: github.com/theislab/sctram

The paper's code and data availability statement is in the Data section.

Tracing map

Proposed by the machine: these links were found in the paper and verified at the source, without human review. The map will receive a Zenodo DOI once one of the paper's authors has validated it with their ORCID.

What the map holds:

5 repositories of the authors' code, each at its verified commit, with its license and how the link was found in the paper;
129 scripts, each with its path and the digest of its content;
19 matches between paragraphs of the paper and lines of the code (method lexical-v1);
neither the text of the paper nor the code itself.

Its JSON (tracing-map.json) is deposited on Zenodo with its DOI once the map is validated.

Data

No dataset and no data link were found in the paper.

Data and code availability

Data: Raw de novo whole embryo data will be available on EMBL-EBI ArrayExpress and BioImage Archive. All data accessions are currently private, and will be made available upon publication:

scRNAseq raw FASTQ and counts (embryo F137), E-MTAB-11504

scRNAseq raw FASTQ and counts (embryo F147), E-MTAB-11520

scRNAseq raw FASTQ and counts (embryo F158), E-MTAB-11911

Multiome raw FASTQ and counts (embryos F137, F147, F158), E-MTAB-16757

Spatial transcriptomics Visium raw FASTQs, counts, images (embryo SOB26), E-MTAB-14785

Spatial transcriptomics Xenium raw cell x gene counts post Xenium cell segmentation, images (embryos SOB25, SOB26), BioImage Archive, S-BIAD2878

The HDCA atlas will be accessible via an interactive web portal at https://cellatlas.io/studies/hdca/. This webportal is currently private and password protected, and will be made available upon publication. Through this portal, users will be able to visualise the dataset and download both raw and processed count matrices, as well as associated metadata and models.

Code: All code will be made publicly available at https://github.com/haniffalab/hdca-paper; the repository is currently private, and will be made available upon publication.

Additional repositories associated with this study include https://github.com/theislab/idtrack181 for biological identifier mapping and harmonization across heterogeneous data sources, and https://github.com/theislab/sctram45, which provides the trajectory-fidelity benchmarking framework introduced in this study.

Reproduced under the paper's license (CC BY), from the paper cited above.

Versions

The history of this record: each version stored by the harvester or made by a correction of its authors or of the maintainers of its code, and what changed in its facts. The texts of the paper (its abstract, its availability statements) are not part of it; versions that changed only those are not listed.

Version 1, 28 September 2026: the first record

Recorded: type, journal, dates, 94 authors, 5 funders, 189 references.

Cite

This paper

Webb, S., Rose, A., Xu, C., Steele, L., Kuri, M. A., Stephenson, E., Inecik, K., Jafree, D., Foster, A. R., Basurto-Lozada, D., Chipampe, N.-J., Pournara, A. V., Jacques, M.-A., To, K., Admane, C., Kritikaki, E., Chroscik, M. A., Horsfall, D., Foreman, J., . . . Haniffa, M. (2026). An integrated single cell and spatial omics atlas of human prenatal development. bioRxiv (preprint). https://doi.org/10.64898/2026.03.30.714220

BibTeX

@article{webb2026integrated, author = {Webb, Simone and Rose, Antony and Xu, Chuan and Steele, Lloyd and Kuri, Masami A and Stephenson, Emily and Inecik, Kemal and Jafree, Daniyal and Foster, April R and Basurto-Lozada, Daniela and Chipampe, Nana-Jane and Pournara, Anna V and Jacques, Marc-Antoine and To, Ken and Admane, Chloe and Kritikaki, Efpraxia and Chroscik, Marta A and Horsfall, David and Foreman, Julia and Rademaker, Koen and Karjalainen, Julia and Laddach, Anna and Madad, Shaista and Lawrence, John EG and Kleshchevnikov, Vitalii and Lisgo, Steven and Lee, Jimmy Tsz Hang and Blevinal, Joséphine and Alqahtani, Ahlam and Makarchuk, Stanislaw and Jackson, Jake and Ucuncu, Ekin and Silva, Teresa P and Lorenzi, Valentina and Torabi, Fereshteh and Botting, Rachel A and Roberts, Kenny and Olabi, Bayanne and Chakala, Keerthi P and Dony, Leander and Dall’Aglio, Gabriele and Cujba, Ana-Maria and Whitfield, Holly J and Hale, Christine and Engelbert, Justin and Li, Tong and Prigmore, Elena and Gopee, Nusayhah Hudaa and Mather, Michael W and Talks, Ben and Fei, Lijiang and Pereira, Diana and Moldovan, Corina and Elmentaite, Rasa and Polanski, Krzysztof and Predeus, Alexander V and Mazin, Pasha and Rowe, Vicky and Bayraktar, Semih and Knight-Schrijver, Vincent R and Clatworthy, Menna and Kanemaru, Kazumasa and Cranley, James and Guo, Andrew and Sinha, Sanjay and McDonald, David and Filby, Andrew and Kerwin, Janet and He, Peng and Sequeira, Inês and Vento-Tormo, Roser and Luecken, Malte D. and Chédotal, Alain and Taylor, Deanne and Martin-Almedina, Silvia and Ostergaard, Pia and Behjati, Sam and Robertson, Elizabeth and Wilson, Nicola K and Marioni, John and Firth, Helen V and Wright, Caroline F and Pourquie, Olivier and Gambardella, Laure and Alexandre, Paula and Gottgens, Berthold and Meyer, Kerstin B and Vijayabaskar, MS and Lotfollahi, Mo and Teichmann, Sarah A and Theis, Fabian J and Progatzky, Fränze and Bayraktar, Omer Ali and Haniffa, Muzlifah}, title = {{An integrated single cell and spatial omics atlas of human prenatal development}}, journal = {bioRxiv (preprint)}, year = {2026}, month = apr, publisher = {bioRxiv}, issn = {2692-8205}, doi = {10.64898/2026.03.30.714220}, url = {https://doi.org/10.64898/2026.03.30.714220} }

RIS

TY - JOUR AU - Webb, Simone AU - Rose, Antony AU - Xu, Chuan AU - Steele, Lloyd AU - Kuri, Masami A AU - Stephenson, Emily AU - Inecik, Kemal AU - Jafree, Daniyal AU - Foster, April R AU - Basurto-Lozada, Daniela AU - Chipampe, Nana-Jane AU - Pournara, Anna V AU - Jacques, Marc-Antoine AU - To, Ken AU - Admane, Chloe AU - Kritikaki, Efpraxia AU - Chroscik, Marta A AU - Horsfall, David AU - Foreman, Julia AU - Rademaker, Koen AU - Karjalainen, Julia AU - Laddach, Anna AU - Madad, Shaista AU - Lawrence, John EG AU - Kleshchevnikov, Vitalii AU - Lisgo, Steven AU - Lee, Jimmy Tsz Hang AU - Blevinal, Joséphine AU - Alqahtani, Ahlam AU - Makarchuk, Stanislaw AU - Jackson, Jake AU - Ucuncu, Ekin AU - Silva, Teresa P AU - Lorenzi, Valentina AU - Torabi, Fereshteh AU - Botting, Rachel A AU - Roberts, Kenny AU - Olabi, Bayanne AU - Chakala, Keerthi P AU - Dony, Leander AU - Dall’Aglio, Gabriele AU - Cujba, Ana-Maria AU - Whitfield, Holly J AU - Hale, Christine AU - Engelbert, Justin AU - Li, Tong AU - Prigmore, Elena AU - Gopee, Nusayhah Hudaa AU - Mather, Michael W AU - Talks, Ben AU - Fei, Lijiang AU - Pereira, Diana AU - Moldovan, Corina AU - Elmentaite, Rasa AU - Polanski, Krzysztof AU - Predeus, Alexander V AU - Mazin, Pasha AU - Rowe, Vicky AU - Bayraktar, Semih AU - Knight-Schrijver, Vincent R AU - Clatworthy, Menna AU - Kanemaru, Kazumasa AU - Cranley, James AU - Guo, Andrew AU - Sinha, Sanjay AU - McDonald, David AU - Filby, Andrew AU - Kerwin, Janet AU - He, Peng AU - Sequeira, Inês AU - Vento-Tormo, Roser AU - Luecken, Malte D. AU - Chédotal, Alain AU - Taylor, Deanne AU - Martin-Almedina, Silvia AU - Ostergaard, Pia AU - Behjati, Sam AU - Robertson, Elizabeth AU - Wilson, Nicola K AU - Marioni, John AU - Firth, Helen V AU - Wright, Caroline F AU - Pourquie, Olivier AU - Gambardella, Laure AU - Alexandre, Paula AU - Gottgens, Berthold AU - Meyer, Kerstin B AU - Vijayabaskar, MS AU - Lotfollahi, Mo AU - Teichmann, Sarah A AU - Theis, Fabian J AU - Progatzky, Fränze AU - Bayraktar, Omer Ali AU - Haniffa, Muzlifah TI - An integrated single cell and spatial omics atlas of human prenatal development T2 - bioRxiv (preprint) J2 - bioRxiv PY - 2026 DA - 2026/04/01 SN - 2692-8205 PB - bioRxiv DO - 10.64898/2026.03.30.714220 UR - https://doi.org/10.64898/2026.03.30.714220 ER -

CSL-JSON

{ "id": "10.64898/2026.03.30.714220", "type": "article", "title": "An integrated single cell and spatial omics atlas of human prenatal development", "container-title": "bioRxiv (preprint)", "author": [ { "family": "Webb", "given": "Simone" }, { "family": "Rose", "given": "Antony" }, { "family": "Xu", "given": "Chuan" }, { "family": "Steele", "given": "Lloyd" }, { "family": "Kuri", "given": "Masami A" }, { "family": "Stephenson", "given": "Emily" }, { "family": "Inecik", "given": "Kemal" }, { "family": "Jafree", "given": "Daniyal" }, { "family": "Foster", "given": "April R" }, { "family": "Basurto-Lozada", "given": "Daniela" }, { "family": "Chipampe", "given": "Nana-Jane" }, { "family": "Pournara", "given": "Anna V" }, { "family": "Jacques", "given": "Marc-Antoine" }, { "family": "To", "given": "Ken" }, { "family": "Admane", "given": "Chloe" }, { "family": "Kritikaki", "given": "Efpraxia" }, { "family": "Chroscik", "given": "Marta A" }, { "family": "Horsfall", "given": "David" }, { "family": "Foreman", "given": "Julia" }, { "family": "Rademaker", "given": "Koen" }, { "family": "Karjalainen", "given": "Julia" }, { "family": "Laddach", "given": "Anna" }, { "family": "Madad", "given": "Shaista" }, { "family": "Lawrence", "given": "John EG" }, { "family": "Kleshchevnikov", "given": "Vitalii" }, { "family": "Lisgo", "given": "Steven" }, { "family": "Lee", "given": "Jimmy Tsz Hang" }, { "family": "Blevinal", "given": "Joséphine" }, { "family": "Alqahtani", "given": "Ahlam" }, { "family": "Makarchuk", "given": "Stanislaw" }, { "family": "Jackson", "given": "Jake" }, { "family": "Ucuncu", "given": "Ekin" }, { "family": "Silva", "given": "Teresa P" }, { "family": "Lorenzi", "given": "Valentina" }, { "family": "Torabi", "given": "Fereshteh" }, { "family": "Botting", "given": "Rachel A" }, { "family": "Roberts", "given": "Kenny" }, { "family": "Olabi", "given": "Bayanne" }, { "family": "Chakala", "given": "Keerthi P" }, { "family": "Dony", "given": "Leander" }, { "family": "Dall’Aglio", "given": "Gabriele" }, { "family": "Cujba", "given": "Ana-Maria" }, { "family": "Whitfield", "given": "Holly J" }, { "family": "Hale", "given": "Christine" }, { "family": "Engelbert", "given": "Justin" }, { "family": "Li", "given": "Tong" }, { "family": "Prigmore", "given": "Elena" }, { "family": "Gopee", "given": "Nusayhah Hudaa" }, { "family": "Mather", "given": "Michael W" }, { "family": "Talks", "given": "Ben" }, { "family": "Fei", "given": "Lijiang" }, { "family": "Pereira", "given": "Diana" }, { "family": "Moldovan", "given": "Corina" }, { "family": "Elmentaite", "given": "Rasa" }, { "family": "Polanski", "given": "Krzysztof" }, { "family": "Predeus", "given": "Alexander V" }, { "family": "Mazin", "given": "Pasha" }, { "family": "Rowe", "given": "Vicky" }, { "family": "Bayraktar", "given": "Semih" }, { "family": "Knight-Schrijver", "given": "Vincent R" }, { "family": "Clatworthy", "given": "Menna" }, { "family": "Kanemaru", "given": "Kazumasa" }, { "family": "Cranley", "given": "James" }, { "family": "Guo", "given": "Andrew" }, { "family": "Sinha", "given": "Sanjay" }, { "family": "McDonald", "given": "David" }, { "family": "Filby", "given": "Andrew" }, { "family": "Kerwin", "given": "Janet" }, { "family": "He", "given": "Peng" }, { "family": "Sequeira", "given": "Inês" }, { "family": "Vento-Tormo", "given": "Roser" }, { "family": "Luecken", "given": "Malte D." }, { "family": "Chédotal", "given": "Alain" }, { "family": "Taylor", "given": "Deanne" }, { "family": "Martin-Almedina", "given": "Silvia" }, { "family": "Ostergaard", "given": "Pia" }, { "family": "Behjati", "given": "Sam" }, { "family": "Robertson", "given": "Elizabeth" }, { "family": "Wilson", "given": "Nicola K" }, { "family": "Marioni", "given": "John" }, { "family": "Firth", "given": "Helen V" }, { "family": "Wright", "given": "Caroline F" }, { "family": "Pourquie", "given": "Olivier" }, { "family": "Gambardella", "given": "Laure" }, { "family": "Alexandre", "given": "Paula" }, { "family": "Gottgens", "given": "Berthold" }, { "family": "Meyer", "given": "Kerstin B" }, { "family": "Vijayabaskar", "given": "MS" }, { "family": "Lotfollahi", "given": "Mo" }, { "family": "Teichmann", "given": "Sarah A" }, { "family": "Theis", "given": "Fabian J" }, { "family": "Progatzky", "given": "Fränze" }, { "family": "Bayraktar", "given": "Omer Ali" }, { "family": "Haniffa", "given": "Muzlifah" } ], "container-title-short": "bioRxiv", "DOI": "10.64898/2026.03.30.714220", "ISSN": "2692-8205", "publisher": "bioRxiv", "URL": "https://doi.org/10.64898/2026.03.30.714220", "issued": { "date-parts": [ [ 2026, 4, 1 ] ] } }

The tracing map gets a citation of its own once an author has validated it and it has a DOI.

Contribute

The authors of this paper can claim it, correct its record and validate its tracing map, and the maintainers of its code (its owner, or a public member of its organization) correct what it says of their repository; anyone signed in can ask for its removal. Every request goes to OSCR's own machine, which answers it; your account page follows them.

Sign in with ORCID to claim this paper as one of its authors, correct its record or validate its tracing map: when the paper's metadata lists your ORCID iD, you are recognized at once. Maintainers of its code: sign in with GitHub, then claim the repository on your account page.

Request its removal

To ask OSCR to remove this record, the copies of its authors' scripts or its tracing map, use the removal request page: signed in, you say who you are, what to remove and why, then review and confirm the request. Published rules decide every request (how).

Discussion, reproductions, activity

Discussion: questions and error reports about this paper and its code, from signed-in readers and its authors. It opens with sign-in.

Reproductions: reports from readers who ran the authors' code: what they reproduced, with which environment, commit and data. It opens with sign-in.

Activity: what happens around this paper: new versions of its record, its map's validation, discussions and reproductions. It opens with sign-in.

An integrated single cell and spatial omics atlas of human prenatal development

Code ↔ Paper

Paper

The authors' code

Overview

Abstract

Repositories

haniffalab/hdca-paper

theislab/idtrack

cellgeni/OMERO_tools

haniffalab/cherita-react

theislab/sctram

Tracing map

Data

Data and code availability

Versions

Version 1, 28 September 2026: the first record

Cite

This paper

Similar papers

Contribute

Claim this paper

Correct its record

Validate its tracing map

Add the badge to its README

Request its removal

Discussion, reproductions, activity