Definition Shapes Discovery

The Core Thesis

Every spatial niche method makes an implicit choice about what a niche is. That choice determines what the method can discover.

This is not a flaw — it is the nature of any measurement. A microscope reveals what its resolution can see. A niche method reveals what its definition can capture.

The problem is that most papers do not make this choice explicit. A method paper says "we identify cellular niches" without specifying which of several possible niche definitions it uses. A benchmark paper compares methods without acknowledging that they define the target concept differently. A review paper lists methods as if they are solving the same problem when they are answering different questions.

The Evidence

Consider three methods applied to the same colorectal cancer tissue:

CellCharter (composition-based) finds 8 neighborhoods defined by cell-type proportions. The "tumor-immune interface" neighborhood contains 40% tumor cells, 25% CD8 T cells, 20% macrophages, 15% other. Every location with this composition gets the same label.

BANKSY (expression-based) subdivides that interface into 3 sub-regions based on expression gradients. The tumor side has proliferative markers. The immune side has activation markers. The boundary itself has a unique gradient signature. These are invisible to composition methods.

NicheCompass (communication-based) reveals that only a fraction of the interface has active PD-L1/PD-1 signaling. The rest has active CXCL9/CXCR3 signaling. Communication-based, these are two completely different niches. Composition-based, they are the same.

Which is "correct"? All of them. They are answering different questions:

• CellCharter answers: Where do these cell types co-localize?
• BANKSY answers: Where does expression change?
• NicheCompass answers: Where are cells communicating?

Implications for Research

For Method Developers

State your niche definition explicitly. Not "we identify niches" but "we define niche as [specific definition] and identify instances of this definition." This makes the method's scope clear and its limitations honest.

For Benchmarkers

Do not compare methods with different niche definitions on the same ground truth. A composition-based and a communication-based method are not solving the same problem. Fair comparison requires either:

1. Methods with the same niche definition compared on appropriate ground truth, or
2. Cross-definition comparison with biological validation (do the different niches predict the same biological outcomes?).

For Users

Match your niche definition to your biological question. If you care about cell-type neighborhoods, use composition methods. If you care about signaling, use communication methods. If you are exploring, try multiple definitions and look for convergent structure — niches identified by multiple definitions are likely biologically robust.

The Meta-Insight

The diversity of niche definitions is not a problem to be solved — it is a reflection of the genuine complexity of tissue microenvironments. A niche is not a single thing. It is a cell-type neighborhood AND an expression microenvironment AND a signaling context AND a morphological region. Different methods illuminate different facets of this multidimensional reality.

The field's maturation will come not from converging on a single definition, but from understanding which definition is right for which question — and from methods that integrate multiple definitions into a unified framework (as NicheCompass begins to do).