Kalin Nonchev

Efficient and accurate search in petabase-scale sequence repositories - Nature MetaGraph enables scalable indexing of large sets of DNA, RNA or protein sequences using annotated de Bruijn graphs.

After years of research and continuous refinement, we’re thrilled to share that our paper on the MetaGraph framework — enabling Petabase-scale search across sequencing data — has been published today in Nature (www.nature.com/articles/s41...)

This project, based on Glib Manaiev’s Master’s thesis, was carried out in close partnership between the Biomedical Informatics Group at @ethz.ch (Gunnar Rätsch @gxxxr.bsky.social), the Computational and Translational Pathology Lab at @UZH.ch and the @unibas.ch (Viktor H. Koelzer).

DeepSpot2Cell: Predicting Virtual Single-Cell Spatial Transcriptomics from H&E images using Spot-Level Supervision Spot-based spatial transcriptomics (ST) technologies like 10x Visium quantify genome-wide gene expression and preserve spatial tissue organization. However, their coarse spot-level resolution aggregat...

Preprint: www.biorxiv.org/content/10.1...
Code: github.com/ratschlab/De...

🎉 DeepSpot2Cell will be presented at NeurIPS 2025 Imageomics!

The idea: model each spot as a bag of cells. 🧬

DeepSpot2Cell combines pathology foundation models + DeepSets neural networks to extract single-cell–level insights from spot data—keeping past experiments relevant and enabling precise cellular analyses.

Older spot-level spatial transcriptomics datasets shouldn't be forgotten now that new single-cell methods exist. 🧬

Instead of discarding this rich resource, we can bridge the gap.
DeepSpot2Cell helps bridge the gap 👇

DeepSpot2Cell: Predicting Virtual Single-Cell Spatial Transcriptomics from H&E images using Spot-Level Supervision

Figure 1

Figure 3

Figure 2

DeepSpot2Cell: Predicting Virtual Single-Cell Spatial Transcriptomics from H&E images using Spot-Level Supervision [new]
Pred. sc gene expr. via DeepSet & spot sup. for spatial transcriptomics.

PS: We also have exciting MSc thesis and semester projects bmi.inf.ethz.ch/opportunitie...

✉️Full job description and how to apply: bmi.inf.ethz.ch/opportunitie...
Application
❗️Applications will be considered only if submitted through the specified process, and incomplete applications will not be considered.

Join us for an exciting internship where cutting-edge machine learning research meets real-world biomedical data!
📍Biomedical Informatics Group of Prof. Gunnar Rätsch @gxxxr.bsky.social, ETH Zürich, Switzerland
⏰ Start: ASAP, full time
💼 Completed PhD in Machine Learning or relevant experience

Internship Opportunity: Multimodal AI Research Scientist at the Biomedical Informatics Group at ETH Zurich 🚀

Interested in working at the intersection of computational pathology, spatial transcriptomics, LLM representation learning, and tissue generation?

Just presented our new multimodal histopathology method "SpotWhisperer" at ICML, one of the largest AI conference.

SpotWhisperer enables spatially resolved annotation of histopathology images using natural language. We achieved this by "transferring" annotations from transcriptomic data. More soon!

🤝 Great collaboration between @bocklab.bsky.social (@moritzbaio.bsky.social, Animesh, Jake), @nonchev.bsky.social, @gxxxr.bsky.social, and pathologist Viktor Kölzer.

SpotWhisperer is at #ICML25 FM4LS workshop. Visit our poster on Saturday (19 July 2025) if you're interested & attending ICML. (6/6)

🔬 Toward histopathology 2.0: spatial transcriptomes inferred from routine diagnostic H&E images + a chat interface for cell-resolution histopathology through English language. (1/6)

Excited to share an update to D3 (DNA Discrete Diffusion) — an application of score-entropy discrete diffusion model for regulatory genomics!

🧬 Paper: biorxiv.org/content/10.110…

(See thread below 👇) (1/n)

#pathology #spatialtranscriptomics #machinelearning

nonchev/TCGA_digital_spatial_transcriptomics · Datasets at Hugging Face We’re on a journey to advance and democratize artificial intelligence through open source and open science.

Explore the dataset: huggingface.co/datasets/non...
Manuscript: www.medrxiv.org/content/10.1...
GitHub: github.com/ratschlab/De...

💡 This resource unlocks exciting opportunities for developing new multi-modal deep learning methods, benchmarking existing ones, and accelerating biological discoveries in cancer research using digital spatial transcriptomics.

🚀 Excited to share that we've generated the largest digital spatial transcriptomics dataset using DeepSpot - over 56 million spatial transcriptomics spots from 3 780 TCGA samples across skin melanoma, renal cell carcinoma, lung adenocarcinoma, and lung squamous cell carcinoma cohorts. #pathology

Glad that you find it exciting too!

The DeepSpot project was carried out in close partnership between the Biomedical Informatics Group at ETH Zurich @gxxxr.bsky.social, the Computational and Translational Pathology Lab at UZH
and @unibas.ch, and the Silina Group at the Institute of Pharmaceutical Sciences, @ethzurich.bsky.social

YouTube video by Broad Institute Autoimmune Disease Machine Learning Challenge Overview

More about the competition: www.youtube.com/watch?v=GUXi...

Leaderboard: hub.crunchdao.com/competitions...

DeepSpot: Leveraging Spatial Context for Enhanced Spatial Transcriptomics Prediction from H&E Images Spatial transcriptomics technology remains resource-intensive and unlikely to be routinely adopted for patient care soon. This hinders the development of novel precision medicine solutions and, more i...

Learn more about DeepSpot, developed at @ethzurich.bsky.social: www.medrxiv.org/content/10.1...

By extending our recent deep learning method, DeepSpot, to support 10x Genomics Xenium data, we significantly improved single-cell gene expression predictions in patients with Inflammatory Bowel Disease. It is exciting to see its performance validated in an independent evaluation!

First place award at the Autoimmune Disease Machine Learning Challenge organized by the @broadinstitute.org and CrunchDAO. Our approach outperformed competitors worldwide in predicting single-cell spatial transcriptomics from H&E images. 🎉

DeepSpot: Leveraging Spatial Context for Enhanced Spatial Transcriptomics Prediction from H&E Images Spatial transcriptomics technology remains resource-intensive and unlikely to be routinely adopted for patient care soon. This hinders the development of novel precision medicine solutions and, more i...

12/12
🔍Read our pre-print at: www.medrxiv.org/content/10.1...
💻Code: github.com/ratschlab/De...
🤗TCGA data: huggingface.co/datasets/non...

11/12 It was carried out in close partnership between the Biomedical Informatics Group at ETH Zurich, the Computational and Translational Pathology Lab at UZH
and @unibas.ch, and the Silina Group at the Institute of Pharmaceutical Sciences, @ethzurich.bsky.social - many thanks!

10/12 This is a joint work with Sebastian Dawo, Karina Selina, Holger Moch, Sonali Andani, Tumor Profiler Consortium, Viktor Hendrik Koelzer, and Gunnar Rätsch 🙌

9/12 The TCGA spatial transcriptomics dataset, containing over 37 million spots, provides unique insights into the molecular landscapes of cancer tissues. It also sets a benchmark for evaluating and developing new spatial transcriptomics models. 🌍

8/12 DeepSpot outperformed previous models or matched bulk-RNA seq performance in tumor type classification. 🧬