Communication-Based Methods¶
Niche definition: active ligand-receptor signaling between spatial neighbors.
These methods answer the question: what are cells saying to each other? They define neighborhoods by intercellular communication rather than co-localization, making them the closest current methods to functional niche definitions.
Key Methods¶
NicheCompass¶
- Paper: Nature Genetics, 2025
- Code: github.com/Lotfollahi-lab/nichecompass
- Niche definition: Graph neural network that explicitly encodes ligand-receptor interactions as features of the spatial graph, learning niche embeddings that reflect both cell identity and intercellular signaling.
- Key innovation: Builds a unified niche atlas across samples and conditions, enabling cross-sample niche comparison with communication-aware embeddings.
- Strengths: Combines the best of composition-based (who is nearby) and communication-based (what are they saying) approaches. Cross-sample integration.
- Limitations: Depends on the L-R database used; niche definitions are constrained by known interactions.
- Verdict: The most complete niche method currently available — integrates cell identity, spatial context, and signaling into a single framework.
COMMOT¶
- Paper: Nature Methods, 2022
- Code: github.com/zcang/COMMOT
- Niche definition: Optimal transport framework that infers spatially-constrained cell-cell communication without requiring single-cell resolution.
- Key innovation: Works with spot-level data (Visium) where multiple cells are in each spot — uses optimal transport to handle mixed spots.
- Strengths: Mathematically principled, handles spot-level data, provides directionality of signaling.
- Limitations: Computational cost scales with dataset size; optimal transport can be expensive for large datasets.
CellChat v2¶
- Paper: Nature Communications, 2021 (v2 update 2024)
- Code: github.com/jinworks/CellChat
- Niche definition: Infers intercellular communication networks from expression data using curated L-R databases with spatial extensions.
- Key innovation: Most widely used cell-cell communication tool; v2 adds spatial awareness and comparison across conditions.
- Strengths: Large user community, extensive L-R database (CellChatDB), rich visualization, condition comparison.
- Limitations: Originally designed for non-spatial scRNA-seq; spatial extensions are added rather than native.
SpaTalk¶
- Paper: Nature Communications, 2022
- Code: github.com/ZJUFanLab/SpaTalk
- Niche definition: Graph network modeling of spatially resolved L-R interactions at single-cell resolution.
- Key innovation: Explicitly designed for single-cell resolution spatial data; models the spatial proximity constraint on L-R interactions.
- Strengths: Single-cell resolution, integrates downstream signaling pathway activity.
SpatialDM¶
- Paper: Nature Communications, 2023
- Code: github.com/StatBiomed/SpatialDM
- Niche definition: Statistical testing for spatially co-expressed ligand-receptor pairs using bivariate Moran's I.
- Key innovation: Provides both global (tissue-wide) and local (per-spot) tests for spatial co-expression of L-R pairs.
- Strengths: Statistical rigor with p-values, distinguishes local hotspots from global trends.
NicheNet¶
- Paper: Nature Methods, 2020
- Code: github.com/saeyslab/nichenetr
- Niche definition: Predicts which ligands from sender cells drive gene expression changes in receiver cells by integrating prior knowledge of signaling and gene regulatory networks.
- Key innovation: Focuses on the downstream effect of signaling rather than just L-R co-expression. multinichenetr extends to multi-sample, multi-condition spatial designs.
- Strengths: Prior knowledge integration, focuses on functional impact of signaling.
- Limitations: Heavily dependent on prior knowledge quality; original version not spatially aware (multinichenetr adds spatial context).
When to Use Communication-Based Methods¶
Best for:
- Understanding intercellular signaling in spatial context.
- Identifying functionally relevant niches defined by active communication rather than just proximity.
- Comparing signaling networks across conditions (e.g., responder vs non-responder).
Not ideal for:
- Tissues where the main spatial organization is driven by composition rather than signaling.
- Datasets with few measured genes that may not cover key L-R pairs.
- When you want niche definitions independent of L-R database completeness.