Deep Reads

In-depth analyses of the most important papers in spatial niche analysis. Each deep read follows the same structure:

1. One-sentence verdict — the honest take.
2. Citation — full reference.
3. Problem setup — what question the paper asks.
4. Niche definition — how this paper defines "niche" (the core lens of this site).
5. Architecture / Method — how it works.
6. Evaluation — what was tested and how.
7. Honest assessment — strengths, limitations, and what the paper does not say.

The Papers

Origin Story

• Schurch et al. 2020 — The paper that started computational niche analysis: 9 cellular neighborhoods in colorectal cancer via CODEX.

Composition-Based

• CellCharter — GNN clustering of spatial neighborhoods across resolutions and platforms.
• CytoCommunity — Supervised niche discovery using clinical condition labels.

Expression-Based

• BANKSY — The gradient feature that captures tissue boundaries.

Covariance-Based

• ENVI/COVET — Niche as covariance tensor, not just mean expression.

Communication-Based

• NicheCompass — Unified niche atlas with ligand-receptor features.
• COMMOT — Optimal transport for spatial cell-cell communication.

Morphology-Based

• TESLA — H&E-guided super-resolution spatial annotation.

Foundation Models

• Nicheformer — Architecture beats scale for spatial tasks.

Downstream

• Niche-DE — Testing whether gene expression depends on niche context.