Claude Skill MartBrowse skillsWhat is a Skill?
 
← Back to list

DeepSpot-M Virtual Spatial Transcriptomics

Predict spatial expression for ~19k protein-coding genes directly from 224x224 H&E histology tiles.

Data & AnalyticsAdvanced★ 33,030⑂ 3,248AI score 8/10Last updated: Aug 9, 2026

What it does

DeepSpot-M is a multimodal foundation model that maps a single 224x224 H&E tile (~20x, ~0.5 µm/px) to gene expression in log1p-CPM. This skill teaches Claude to:

  • Install deepspotm, request the gated Hugging Face weights, and authenticate
  • Insert tile-size/magnification guards so mismatched crops never reach the model
  • Call predict_genes with a list of HGNC symbols and keep output columns labelled
  • Choose among five frozen gene-embedding sources (evo2, orthrus, prott5, scgpt, apertus)
  • Run whole-slide workflows: histolab tiling → batched inference → tiles-by-genes matrix → AnnData

Because genes enter as queryable embeddings rather than fixed output slots, coverage goes far beyond the few hundred genes of a typical spatial assay panel and includes genes unseen in training.

Who it's for

  • Researchers with abundant histology slides but no matching spatial transcriptomics assay
  • Computational pathology teams adding an expression channel to large cohorts like TCGA
  • Developers extending a morphology-only pipeline with predicted gene signals
  • Noncommercial research only (code: PolyForm NC; weights: CC-BY-NC-SA-4.0)

Examples

  1. Marker gene maps: tile a tumour section and query ["EPCAM", "CD3D"] to render epithelial vs. immune spatial heatmaps.
  2. Cohort atlas: batch-infer hundreds of slides, store tiles-by-genes matrices with coordinates, and assemble them into AnnData for downstream spatial analysis.
  3. Embedding-source sanity check: run the same tiles through scgpt and prott5 to confirm findings are robust to the gene-representation choice.

· · · Install guide · · ·

Install in the Claude app (no terminal)
  1. Download the ZIP with the button below.
  2. In Claude, open Settings → Capabilities and turn on 'Code execution and file creation'. (one time)
  3. Go to Customize → Skills → + → 'Upload a skill' and upload the ZIP.
↓ Download ZIP
Install in Claude Code

Let Claude do it — paste this into Claude Code

Install the skill I found on Claude Skill Mart.
Copy the skills/deepspot-m folder from the GitHub repo K-Dense-AI/scientific-agent-skills into my ~/.claude/skills/deepspot-m/.
When it's done, tell me in one line what this skill can do.

Install with a command instead

git clone https://github.com/K-Dense-AI/scientific-agent-skills.git && mkdir -p ~/.claude/skills && cp -r scientific-agent-skills/skills/deepspot-m ~/.claude/skills/

⚠ This is a third-party skill. Check the source repository before installing.

  1. Clone the repository: git clone https://github.com/K-Dense-AI/scientific-agent-skills.git
  2. Copy the skill folder into Claude: mkdir -p ~/.claude/skills && cp -r scientific-agent-skills/skills/deepspot-m ~/.claude/skills/
  3. Prepare a Python 3.10–3.13 environment; if you want GPU inference, install the PyTorch build matching your CUDA version first.
  4. Install the model package: uv pip install deepspotm==1.0.0 (or pip install deepspotm==1.0.0).
  5. Visit https://huggingface.co/ratschlab/DeepSpotM and request access to the gated weights, then wait for approval.
  6. After approval, run huggingface-cli login on the machine that will download the weights (once per machine).
  7. Restart Claude Code and try a prompt like "Predict spatial gene expression from these H&E tiles with DeepSpot-M".
  8. Review the PolyForm Noncommercial and CC-BY-NC-SA terms before publishing or redistributing any outputs.
View source on GitHub ↗License: MIT