Tutorials

Short, task-oriented walkthroughs.
Use these pages when you want an end-to-end example rather than a conceptual overview.

---

Quickstart

Choose the fastest path for your environment:

• Python / Scanpy (Colab): minimal end-to-end run on a public dataset
  → Open notebook on GitHub

• OpenAI backend (Colab): same workflow as the Python / Scanpy quickstart, but configured for the OpenAI backend. OPENAI_API_KEY requires OpenAI API billing (paid API credits).
  → Open notebook on GitHub

• R / Seurat (Colab): Seurat-oriented quickstart with an export → Python run → re-import pattern
  → Open notebook on GitHub

---

Advanced

• R via reticulate: run LLM-scCurator from R through a Python bridge (advanced usage)
  → Open notebook on GitHub

• Spatial validation (paper): manuscript notebooks for Xenium/Visium spatial plots and pseudo-bulk summaries
  → Colon Xenium (Fig.2h; ED Fig.3d): Open notebook on GitHub
  → OSCC Visium (ED Fig.3a–c): Open notebook on GitHub

• Fully local LLM (Ollama): Curate features and optionally annotate clusters using a local LLM backend (no external transmission).
  → Open notebook on GitHub

• Local feature distillation → Approved chat LLM annotation (no external LLM API calls): Curate features locally, export a curated cluster→genes table, then annotate it via an institution-approved chat interface (e.g., Microsoft Copilot “Work”) by uploading the CSV/Excel or pasting markers.
  → Open notebook on GitHub

---

Notes

• For the underlying design (masking, rescue, leakage filtering, hierarchy), see Concepts.
• For deterministic reproduction (benchmarks/figures/Source Data), follow paper/README.md.