stat-agent 0.1.1

STAT: Spatial Transcriptomics Analytical agenT - AI-powered platform for spatial omics analysis

STAT

Spatial Transcriptomics Analytical agenT

An AI-powered platform for spatial omics analysis with multi-format support, interactive visualization, and intelligent code generation.

Features

• AI Agent: Natural language interface for spatial transcriptomics analysis — ask questions, get results
• Multi-format support: Single-slice, multi-slice, and multi-omics (gene + protein) datasets
• Interactive viewer: Canvas-based spatial visualization with zoom/pan, ROI drawing, and cell overlays
• Skill system: Extensible analysis skills (cell type annotation, deconvolution, spatial domains, etc.)
• Code execution: Agent generates and runs analysis code in a sandboxed environment
• Multi-provider LLM: Works with OpenAI, Anthropic, Google, Deepseek, and Poe

Installation

pip install stat-agent

With analysis skill dependencies (squidpy, scvi-tools, etc.):

pip install "stat-agent[skills]"

Quick Start

Web Interface

stat-web
# Open http://localhost:8889

Or with the startup script (includes Jupyter Lab):

./start_web.sh

In the web UI:

1. Enter path to your dataset directory
2. Configure LLM (API key, model)
3. Click "Load Dataset"
4. Ask questions in the chat panel: "Annotate cell types", "Find spatially variable genes", "Show BRCA1 expression"

Data Format

STAT auto-detects your data layout. Place files in a single directory:

Single-slice:

dataset/
├── tissue.h5ad          # Required: AnnData with x, y coordinates in obs
└── he.tif               # Optional: H&E image (pixel coords = cell coords)

Multi-slice:

dataset/
├── tissue_slice_0.h5ad
├── he_slice_0.tif
├── tissue_slice_1.h5ad
└── he_slice_1.tif

Multi-omics:

dataset/
├── tissue.h5ad          # Gene expression
├── tissue_protein.h5ad  # Protein expression
├── he.tif
└── protein_CD3.tif

Key: Cell coordinates (x, y) in adata.obs map directly to image pixels (x, y). No coordinate transformation needed.

Built-in Skills

Skill	Description
Cell Type Annotation (GPT)	Unsupervised clustering + LLM-based annotation
Cell Type Annotation (scANVI)	Transfer learning from scRNA-seq reference
Deconvolution (RCTD)	Spot-level cell type deconvolution
Spatial Domains (SpaGCN)	Graph-based spatial domain identification
SVG (SpatialDE)	Spatially variable gene detection
Neighborhood Enrichment	Cell type co-localization analysis
Cell Communication (LIANA+)	Ligand-receptor interaction analysis
Cell Communication (CellPhoneDB)	Permutation-based interaction testing
GO Enrichment	Gene Ontology pathway analysis
Niche Detection (Harmonics)	Spatial niche identification
Integration (Harmony)	Multi-slice batch correction
Alignment (STalign)	Spatial slice alignment

Architecture

User Query → QueryPlanner → SkillFilter → LLM Matching → SkillVerifier → Code Generation → Execution

• QueryPlanner: Determines target slices, breaks complex queries into steps
• SkillFilter: Programmatic filtering by modality, data level, number of slices
• SkillVerifier: Checks prerequisites, requests missing information
• SpatialAgent: Generates analysis code using skill instructions + session context
• CodeExecutor: Sandboxed execution with state change detection

Project Structure

stat_agent/
├── core/                  # Data layer
│   ├── session.py         # Multi-slice/multi-omics session
│   ├── data_slice.py      # Single data slice wrapper
│   └── roi_manager.py     # ROI geometry management
├── agent/                 # Agent pipeline
│   ├── spatial_agent_core.py
│   ├── conversation_orchestrator.py
│   ├── pipeline_executor.py
│   ├── query_planner.py
│   ├── skill_registry.py
│   ├── skill_filter.py
│   ├── skill_verifier.py
│   ├── llm_backend.py
│   └── memory.py
└── functions/
    └── io.py              # Data loading
.claude/skills/            # Skill definitions (SKILL.md + helper libs)
web_interface.py           # Flask backend + API endpoints
static/                    # Frontend (JS + CSS)
templates/                 # HTML templates

License

BSD-3-Clause