This dataset is released under CC-BY-NC-SA-4.0 strictly for non-commercial academic research. Commercial use is not permitted unless explicit written permission is granted by the authors.
ELIGIBILITY. Access is granted only to individuals whose affiliations are exclusively academic or public non-profit research institutions. Anyone holding any concurrent commercial affiliation — employment, consulting, advisory roles, internships, or founding roles at a company or startup — is not eligible, regardless of intended use.
SCOPE. Research performed at, for, funded by, or in collaboration with a commercial entity is commercial use and is prohibited. This includes corporate R&D, internal evaluation, benchmarking, and proof-of-concept work, even if exploratory and never productized. "Research" conducted in a commercial setting does not qualify as non-commercial academic research.
REVIEW. All requests are reviewed manually against these criteria; please allow 3-5 business days. By submitting, you agree to share your contact information (email and username) with the repository authors. Vague, incomplete, or inaccurate submissions will be declined. False or incomplete declarations void any access granted and constitute a violation of the license terms. For commercial licensing, contact the maintainers.

KPMP virtual spatial transcriptomics

Virtual spatial transcriptomics for 1,534 H&E whole-slide images (fresh-frozen and FFPE) of kidney biopsies from 476 participants of the Kidney Precision Medicine Project (KPMP): 19,338 genes in each of 5,458,825 spots on a 56 µm grid, predicted with DeepSpot-M Kidney.

Model: DeepSpot-M Kidney, finetuned from ratschlab/DeepSpotM · Code: github.com/ratschlab/DeepSpotM · Paper: citation will be added on publication; DeepSpot-M: medRxiv 2026.06.19.26356060.

News

[09.2026] Introducing Aurora - a no-code platform for virtual spatial transcriptomics from H&E images. Try the interactive Aurora demo.

Enrollment group	FF slides	FFPE slides	Participants	Spots
AKI	28	68	33	251,920
CKD	485	819	401	4,779,144
DM-R	53	72	36	387,565
Healthy Reference	–	9	6	40,196
Total	566	968	476	5,458,825

data/<ENROLLMENT_GROUP>/<FF|FFPE>/<slide_id>.h5ad    one AnnData file per slide (144 GB in total)
metadata.csv                                         one row per slide

Example

FFPE biopsy of the AKI group: H&E and virtual LRP2 (proximal tubule), UMOD (thick ascending limb) and COL1A1 (fibrosis and stroma).

More examples (CKD FF, DM-R FFPE, Healthy Reference FFPE)

Download

This is a gated dataset: log in first (pip install huggingface_hub).

import pandas as pd
from huggingface_hub import login, snapshot_download, hf_hub_download
login(token="YOUR_HF_TOKEN")

repo = "ratschlab/KPMP_virtual_spatial_transcriptomics_atlas"

# one enrollment group and slide type, e.g. AKI FFPE ("data/*/FF/*" for all fresh-frozen slides)
snapshot_download(repo, repo_type="dataset", local_dir="KPMP_data", allow_patterns="data/AKI/FFPE/*")

# or select slides in metadata.csv first
meta = pd.read_csv(hf_hub_download(repo, "metadata.csv", repo_type="dataset"), low_memory=False)
sel = meta[meta["Primary Adjudicated Category"] == "Diabetic Kidney Disease"]
snapshot_download(repo, repo_type="dataset", local_dir="KPMP_data", allow_patterns=sel.file_path.tolist())

Usage

import numpy as np
import scanpy as sc
import squidpy as sq

adata = sc.read_h5ad("KPMP_data/data/AKI/FFPE/ba1203a0-db6f-43ec-a80a-ebb85c2ee193_S-2406-007749_HE_4of4.h5ad")
adata.X = adata.X.astype(np.float32)
sq.pl.spatial_scatter(adata, color=[None, "LRP2", "UMOD", "COL1A1"], shape="square", ncols=2)

Data format

X: virtual expression of 19,338 genes per spot, log1p counts per 10,000 (log1p CP10k), stored as float16 (cast to float32 before analysis).
obs: one row per spot: participant_id, enrollment, slide_type, section (tissue piece on the slide, 1 = largest), lattice position (array_row, array_col), centre in full-resolution pixels (x_pixel, y_pixel) and per-spot image QC (qc_*).
var: HGNC gene symbol, ensembl_ids, hgnc_id.
obsm['spatial'] and uns['spatial']: spot centres and an H&E overview image, as scanpy and squidpy expect.
uns['he2st']: slide and processing metadata.
metadata.csv: one row per slide: file path, participant, enrollment group, slide type, number of spots, the KPMP open-access clinical table, the KPMP pathology descriptor scores (TIV: … columns) and the KPMP download link of the whole-slide image (svs_url).

Each spot is a 56 µm square of a non-overlapping lattice over the tissue; its expression is predicted from the 65 µm H&E field of view centred on it. One model trained on KPMP 10x Visium biopsies of other participants was applied to every slide.

How to cite

Please cite DeepSpot-M (the citation of the kidney paper will be added on publication):

@article{nonchev2026deepspotm,
  title={DeepSpot-M: a multimodal foundation model for transcriptome-wide virtual spatial transcriptomics from histology},
  author={Nonchev, Kalin and Dawo, Sebastian and Silina, Karina and Koelzer, Viktor Hendrik and Raetsch, Gunnar},
  journal={medRxiv},
  pages={2026--06},
  year={2026},
  publisher={Cold Spring Harbor Laboratory Press}
}

KPMP acknowledgement

The H&E slides, clinical data and pathology descriptor scores are KPMP open-access data. Work that uses this dataset must acknowledge KPMP (Citing KPMP data):

The results here are in whole or part based upon data generated by the Kidney Precision Medicine Project. Accessed Month Day, Year. https://www.kpmp.org.

The Kidney Precision Medicine Project (KPMP) is supported by the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) through the following grants: U01DK133081, U01DK133091, U01DK133092, U01DK133093, U01DK133095, U01DK133097, U01DK114866, U01DK114908, U01DK133090, U01DK133113, U01DK133766, U01DK133768, U01DK114907, U01DK114920, U01DK114923, U01DK114933, U24DK114886, UH3DK114926, UH3DK114861, UH3DK114915, UH3DK114937, 1U01DK144965-01, 1U01DK144994-01. We gratefully acknowledge the essential contributions of our patient participants and the support of the American public through their tax dollars.

License

CC BY-NC-SA 4.0, non-commercial academic research only. The KPMP data included remain subject to KPMP's terms of use.

NB

Computational data analysis was performed at Leonhard Med, the secure trusted research environment at ETH Zurich. Our pipeline aligns with the specific cluster requirements and resources.

Downloads last month: 219

Total file size:

144 GB

Eligibility declarations required to request access