Why AI Agents for Spatial Transcriptomics?
Spatial transcriptomics is one of the fastest-growing fields in genomics. Technologies like 10x Visium, Xenium, MERFISH, and Slide-seq generate rich datasets that combine gene expression with spatial location — but analyzing them requires chaining together multiple specialized tools (Scanpy, SpaGCN, STAGATE, Cell2location, CellPhoneDB...).
AI agents can orchestrate these tools through natural language, remembering your data and preferences across sessions. Instead of memorizing CLI commands, you describe what you want.
Available AI Tools
OmicsClaw — 15 Spatial Skills with Memory
OmicsClaw offers 15 spatial transcriptomics skills with a persistent memory system:
| Skill | What It Does |
|---|---|
spatial-preprocessing | QC, normalization, HVG, PCA, UMAP, clustering |
spatial-domain-identification | Tissue region identification (Leiden, SpaGCN, STAGATE, BANKSY) |
spatial-cell-annotation | Cell type annotation (marker-based, Tangram, scANVI) |
spatial-deconvolution | Cell type proportions (Cell2location, RCTD, SPOTlight) |
spatial-svg-detection | Spatially variable genes (Moran's I, SpatialDE, SPARK-X) |
spatial-cell-communication | Ligand-receptor interactions (LIANA+, CellPhoneDB, CellChat) |
spatial-velocity | RNA velocity (scVelo, VELOVI) |
spatial-trajectory | Developmental trajectories (CellRank, Palantir) |
spatial-cnv | Copy number variation (inferCNVpy) |
Key advantage: memory. OmicsClaw remembers your datasets across sessions. Upload a Visium file on Monday, come back Wednesday, and it knows your data, previous analyses, and preferred parameters.
K-Dense Scientific Skills
K-Dense includes bioinformatics skills connected to 250+ databases. For spatial work, use the genomics and bioinformatics skill sets.
LabClaw — Stanford-Princeton Skills
LabClaw includes 86 biology skills with tooluniverse-spatial-transcriptomics for spatial analysis, plus scanpy and anndata skills for the underlying data manipulation.
Example: Full Spatial Analysis Pipeline
Using OmicsClaw via Telegram:
Session 1:
You: [Upload visium_brain.h5ad]
You: "Preprocess this Visium data"
Bot: Running QC, normalization, clustering...
Found 5,000 spots, 20,000 genes. Filtered to 18,500 HVGs.
UMAP and leiden clustering complete (12 clusters).
[Saves to memory]
You: "Find spatial domains"
Bot: Running domain identification with SpaGCN and STAGATE...
Identified 8 spatial domains.
[Sends UMAP + spatial domain overlay plots]
Session 2 (next day):
You: "What spatially variable genes are in domain 3?"
Bot: Using your Visium data from yesterday (5,000 spots, 8 domains).
Running SVG detection on domain 3...
Top 10 SVGs: GFAP (p=1.2e-15), MBP (p=3.4e-12), ...Choosing the Right Tool
| Need | Tool | Why |
|---|---|---|
| Full spatial pipeline with memory | OmicsClaw | 15 dedicated spatial skills, remembers data across sessions |
| Quick one-off analysis | K-Dense skills | Connect to Claude Code, use bioinformatics skills directly |
| Lab-wide deployment | LabClaw + OpenClaw | 240 skills including spatial, team collaboration |
| Publication figures | OmicsClaw + matplotlib | Standardized output with figures/ directory |
Supported Platforms
| Platform | Format | Skills Support |
|---|---|---|
| 10x Visium | .h5ad | Full (OmicsClaw, K-Dense, LabClaw) |
| 10x Xenium | .h5ad | Full |
| MERFISH | .h5ad | Full |
| Slide-seq | .h5ad | Full |
| 10x scRNA-seq | .h5ad | Single-cell skills |
Getting Started
-
Install OmicsClaw for the most complete spatial experience:
git clone https://github.com/TianGzlab/OmicsClaw.git cd OmicsClaw pip install -e ".[spatial]" -
Or use K-Dense skills with Claude Code for quick analysis:
git clone https://github.com/K-Dense-AI/claude-scientific-skills.git -
Prepare your
.h5adfile and start chatting with your data.
Links
- OmicsClaw
- K-Dense Scientific Skills
- LabClaw
- Claw4Science — explore the full ecosystem