Archetype Crosstalk Network

Adapted from [AMG+23]

Packages

To run this notebook, install ParTIpy with the tutorial extras: pip install partipy[extra].

import scanpy as sc
import liana as li
import pandas as pd

import partipy as pt
from partipy.datasets import load_fibroblast_data
from partipy.crosstalk import get_specific_genes_per_archetype, get_archetype_crosstalk, plot_weighted_network

Data

Here we use the data from [MGenoveL+20].

adata = load_fibroblast_data(
    verbose=True,
)
adata

Loading processed data from data/fibroblast_muhl_processed.h5ad

AnnData object with n_obs × n_vars = 1374 × 17790

sc.pp.normalize_total(adata)
sc.pp.log1p(adata)
sc.pp.highly_variable_genes(adata)
sc.pp.pca(adata, mask_var="highly_variable", n_comps=50)
adata.layers["z_scaled"] = sc.pp.scale(adata.X, max_value=10, copy=True)  # save this for later
sc.pl.pca_scatter(adata, color=["Bmp7", "Wnt2"])

Based on the PCA scree plot, we will be using 8 PC dimensions to fit the archetypes.

sc.pl.pca_variance_ratio(adata, n_pcs=50)
pt.set_obsm(adata=adata, obsm_key="X_pca", n_dimensions=8)

pt.compute_selection_metrics(adata=adata, n_archetypes_list=range(2, 10))
pt.plot_var_explained(adata).show()

pt.plot_IC(adata).show()

pt.compute_bootstrap_variance(adata, n_bootstrap=30, n_archetypes_list=range(2, 10))
pt.plot_bootstrap_variance(adata)

pt.plot_bootstrap_2D(adata, result_filters={"n_archetypes": 6})

Here, we will use 5 archetypes. When fitting 4 archetypes we have observed that running archetypal analysis on bootstrapped data shows that one archetype is quite unstable, whereas using 5 worked much better.

pt.plot_archetypes_2D(adata=adata, show_contours=True, result_filters={"n_archetypes": 6})

Determine which genes are characteristic for each archetype

pt.compute_archetype_weights(adata=adata, mode="automatic", result_filters={"n_archetypes": 6})
pseudobulk = pt.compute_archetype_expression(adata=adata, layer="z_scaled", result_filters={"n_archetypes": 6})

arch_idx = 2
adata.obs[f"weights_archetype_{arch_idx}"] = pt.get_aa_cell_weights(adata, n_archetypes=6)[:, arch_idx]
pt.plot_archetypes_2D(adata=adata, color=f"weights_archetype_{arch_idx}", result_filters={"n_archetypes": 6})

Applied length scale is 12.34.

specific_genes_per_archetype = get_specific_genes_per_archetype(archetype_expression=pseudobulk, min_score=0.05)

[(arch_id, len(arch_genes)) for arch_id, arch_genes in specific_genes_per_archetype.items()]

[(0, 12), (1, 1173), (2, 1722), (3, 760), (4, 4871), (5, 1381)]

Based on the genes we infer the functions of each archetype

for i in range(len(specific_genes_per_archetype)):
    print(i, specific_genes_per_archetype[i].head(25)["gene"].to_list())

0 ['Serping1', 'Bcl6', 'Zfp808', 'Itm2b', 'Ms4a4d', 'Rax', 'C1rb', 'Gm2237', 'Hoxc10', 'Gm7670', 'Gm23119', 'Cyp1b1']
1 ['Lmod1', 'Cnn1', 'Actg2', 'Cdh6', 'Pdgfc', 'Myocd', 'Itga8', 'Asb2', 'Myh11', 'Trpc6', 'Kcnmb1', 'Tnfrsf19', 'Tspan2', 'Acta1', 'Tagln', 'Pcp4l1', 'Slc2a4', 'Hspb7', 'Gm16000', 'Pde6h', 'Ctxn1', 'Hhip', 'Ptk2b', 'Cacna1h', 'Sntg2']
2 ['Nrg1', 'Cd38', '1700071M16Rik', 'Lef1', 'Lpar3', 'Lef1os1', 'Trpa1', 'Pla2g7', 'Lama5', 'Cdhr17', 'Rab3b', 'Ano2', 'Ptprr', 'C5ar1', 'Glp2r', 'Cd300e', 'Abcc4', 'Tbx3', 'Sema4d', 'Paqr5', 'Insc', 'Eva1a', 'Ednrb', 'Procr', 'Ano1']
3 ['Efnb2', 'Gm42721', 'Ugcg', 'Pde1c', 'Gm37515', 'Gm31258', '1700082M22Rik', 'Gm42785', 'Gm37954', 'Ryr3', 'Bmp3', 'Gm43577', 'Gm42513', 'Gm44153', 'Gm4117', 'Kcnq1ot1', 'Macrod2os1', 'Ksr2', 'Gm17131', 'Gm37315', 'Ddx17', 'Mir6240', 'Pkib', 'Gm12018', 'Gm37086']
4 ['Sox18', 'Mmrn2', 'Robo4', 'Ushbp1', 'Tie1', 'Plvap', 'Flt1', 'Kdr', 'Emcn', 'Egfl7', 'Podxl', 'Dll4', 'Cd300lg', 'Depp1', 'Adgrl4', 'Gdf10', 'Fabp4', 'Apold1', 'Gata2', 'Gimap6', 'Fam110d', 'Sox17', 'Mmp3', 'Car4', 'Adgrg1']
5 ['Efhd1os', 'Efhd1', '1700019D03Rik', 'Npy1r', 'Cmah', 'Gm16083', 'Sema3c', 'Adgrg2', 'Dpp4', 'Ebf1', 'Tnfrsf11b', 'Gm12043', 'Mfap5', 'Rab32', 'Cd55', 'Adamts5', 'Fstl1', 'Nhsl1', 'Anxa3', 'Ackr4', 'Tek', 'Ugp2', 'Cadm3', 'Qpct', 'Creb5']

archetype_functions = {
    0: "ECM-remodeling fibroblasts",
    1: "Vascular-associated fibroblasts",
    2: "Neuronal/epithelial-interacting fibroblasts",
    3: "Myofibroblasts",
    4: "Immune-modulatory fibroblasts",
}

Finally we use LIANA+ to infer ligand and receptor are enriched between archetypes.

mouse_resource = li.rs.select_resource("mouseconsensus")
mouse_resource

	ligand	receptor
31371	Dll1	Notch1
31372	Dll1	Notch2
31373	Dll1	Notch4
31374	Dll1	Notch3
31375	Nrg2	Erbb2_Erbb3
...	...	...
35355	Serpina1a	Lrp1
35356	Serpina1b	Lrp1
35357	Serpina1c	Lrp1
35358	Serpina1d	Lrp1
35359	Serpina1e	Lrp1

3989 rows × 2 columns

archetype_crosstalk = get_archetype_crosstalk(archetype_genes=specific_genes_per_archetype, lr_resource=mouse_resource)
for sender_arch in archetype_crosstalk.keys():
    for receiever_arch in archetype_crosstalk[sender_arch].keys():
        print(f"{sender_arch} -> {receiever_arch}: {len(archetype_crosstalk[sender_arch][receiever_arch])}")
    break

0 -> 0: 0
0 -> 1: 0
0 -> 2: 0
0 -> 3: 0
0 -> 4: 1
0 -> 5: 0

The corresponding flattened version looks like this:

flat_df = pd.concat(
    [
        df.assign(sender_archetype=sender_arch, receiver_archetype=receiver_arch)
        for sender_arch, receivers in archetype_crosstalk.items()
        for receiver_arch, df in receivers.items()
    ],
    ignore_index=True,
)
flat_df

	ligand	receptor	ligand_z_score	ligand_specificity_score	receptor_z_score	receptor_specificity_score	sender_archetype	receiver_archetype
0	Serping1	Lrp1	0.205085	0.154098	0.366762	0.080052	0	4
1	Sema5a	Met	0.568509	0.403587	1.126435	0.970279	1	1
2	Lgals1	Itgb1	0.127651	0.062814	0.573087	0.474441	1	1
3	Vegfb	Adrb2	0.282850	0.242285	0.419965	0.250148	1	1
4	Tln1	Itgb3	0.250728	0.074158	0.499605	0.385054	1	1
...	...	...	...	...	...	...	...	...
763	C3	Cd81	0.549103	0.142512	0.564969	0.195734	5	5
764	Adam10	Tspan5	0.156096	0.086548	0.212523	0.063955	5	5
765	Lamc2	Cd151	0.250121	0.133691	0.228000	0.119776	5	5
766	Wnt11	Klrg2	0.232272	0.201278	0.081127	0.103476	5	5
767	Fn1	Dpp4	0.233996	0.115994	0.364511	0.404768	5	5

768 rows × 8 columns

plot_weighted_network(
    specific_genes_per_archetype=specific_genes_per_archetype,
    archetype_crosstalk_dict=archetype_crosstalk,
    layout="shell",
    seed=42,
)

Session Info

import session_info2

print(session_info2.session_info(dependencies=True))

pandas	2.2.3
plotnine	0.14.5
scanpy	1.11.2
liana	1.5.1
partipy	0.2.0 (0.0.1)
anndata	0.11.4
----	----
dask	2024.11.2
natsort	8.4.0
cattrs	24.1.3
pytz	2025.2
crc32c	2.7.1
debugpy	1.8.14
stack-data	0.6.3
mizani	0.13.5
charset-normalizer	3.4.2
wcwidth	0.2.13
psutil	7.0.0
future	1.0.0
certifi	2025.4.26 (2025.04.26)
parso	0.8.4
ipykernel	6.29.5
torch	2.10.0
typing-inspection	0.4.2
cycler	0.12.1
ipython	9.2.0
tornado	6.5.1
narwhals	1.41.0
urllib3	2.4.0
matplotlib	3.10.3
setuptools	80.8.0
idna	3.10
networkx	3.5
scipy	1.15.2
legacy-api-wrap	1.4.1
MarkupSafe	3.0.3
scikit-learn	1.6.1
Jinja2	3.1.6
docrep	0.3.2
pydantic	2.12.0
zstandard	0.25.0
prompt_toolkit	3.0.51
numpy	2.2.6
python-dateutil	2.9.0.post0
appnope	0.1.4
coverage	7.10.7
importlib_metadata	8.7.0
numba	0.61.2
mudata	0.3.2
attrs	25.3.0
matplotlib-inline	0.1.7
toolz	1.0.0
llvmlite	0.44.0
jedi	0.19.2
requests-cache	1.2.1
h5py	3.13.0
pillow	11.2.1
numcodecs	0.15.1
typing_extensions	4.15.0
pyarrow	20.0.0
asttokens	3.0.0
jupyter_client	8.6.3
zarr	2.18.7
requests	2.32.3
wrapt	1.17.2
Deprecated	1.2.18
threadpoolctl	3.6.0
PyYAML	6.0.2
plotly	6.1.2
tqdm	4.67.1
pybiomart	0.2.0
statsmodels	0.14.4
colorama	0.4.6
executing	2.2.0
pure_eval	0.2.3
jupyter_core	5.8.1
pyzmq	26.4.0
asciitree	0.3.3
annotated-types	0.7.0
url-normalize	2.2.1
comm	0.2.2
ipywidgets	8.1.7
platformdirs	4.3.8
decorator	5.2.1
pydantic_core	2.41.1
xarray	2024.11.0
packaging	25.0
kiwisolver	1.4.8
traitlets	5.14.3
zipp	3.22.0
joblib	1.5.1
six	1.17.0
patsy	1.0.1
cloudpickle	3.1.1
session-info2	0.1.2
Pygments	2.19.1
pyparsing	3.2.3
----	----
Python	3.11.0 | packaged by conda-forge | (main, Jan 14 2023, 12:26:40) [Clang 14.0.6 ]
OS	macOS-15.7.1-arm64-arm-64bit
Updated	2026-02-16 20:32