Case Study

Decoding Cell–Cell Communication from Single-Cell RNA-Seq Data

Inquiry

Technical Bulletin · Single-Cell Communication

Decoding Cell–Cell Communication from Single-Cell RNA-Seq Data

How ligand–receptor inference connects sender cells, receiver responses and disease-associated signaling programs across samples and conditions.

Discuss Your Cell Communication Study

The translational question

Patients may share a diagnosis while their tissues run different communication programs.

Single-cell analysis can reveal those programs, but biomarker discovery requires moving from cell-level predictions to patient-level, measurable evidence.

Which cells send the disease-driving signal?Identify source populations, receiver states and communication direction rather than treating a ligand or receptor as an isolated marker.

Does the signal differ between clinically meaningful groups?Estimate communication per sample and compare responders versus non-responders, disease subtypes or longitudinal time points with donor-aware statistics.

Can the interaction be measured outside discovery data?Prioritize axes supported by tissue expression, receptor-positive cell abundance, downstream response and an assay-compatible component.

Will it generalize to a future patient?Evaluate stability across cohorts, sample handling, analytical choices and clinically realistic prevalence—not only significance in a pooled atlas.

Core component 01

What ligand–receptor inference actually connects

Communication inference starts from coordinated expression and curated molecular knowledge. A useful translational analysis retains the complete chain from sender to receiver response and then asks whether that chain varies reproducibly across patients.

Sender cell stateligand expression and secretion context

LLigand → receptor complexcurated prior + expression criteria

Receiver-cell responsereceptor availability, cofactors, pathway activity, transcription-factor activity and ligand-linked target genes

Patient-level phenotypesubtype · response · toxicity · progression

Expression supports possibilityRNA abundance does not prove protein secretion, receptor binding or pathway activation.

Direction mattersThe same pair can have different interpretations depending on sender, receiver and disease stage.

Context creates valueSpatial adjacency, downstream activity and clinical association strengthen a candidate interaction.

Core component 02

Tool selection should follow the biological decision

No single tool is optimal for every communication question. Database coverage, scoring assumptions, multi-subunit complexes, downstream-response modeling, sample replication and spatial information all change what can be inferred. The table is a project-selection guide rather than a performance ranking.

Tool / approach	Best fit	Distinctive strength	Important consideration	Translational use
CellPhoneDB	Pair discovery	Curated human interactions and explicit treatment of multi-subunit ligand–receptor complexes.	Permutation-based cell-group specificity is not a patient-level response model by itself.	Generate interpretable candidate pairs for targeted validation.
CellChat	Network patterns	System-level communication probabilities, pathway aggregation and sender/receiver network roles.	Results depend on cell annotation, database version and comparison strategy.	Identify altered signaling pathways and dominant cellular hubs.
NicheNet	Ligand activity	Links candidate ligands to downstream genes observed in a receiver population.	Requires a well-defined receiver gene set and a biologically meaningful contrast.	Connect extracellular candidates to a response signature or mechanism.
LIANA / LIANA+	Consensus	Unifies methods and resources; supports consensus, multi-condition, spatial and multi-omics analyses.	Consensus reduces method-specific dependence but does not remove data or prior-knowledge limitations.	Build robust, provenance-aware interaction shortlists across analysis strategies.
MultiNicheNet	Multi-sample contrast	Designed for differential communication across multiple samples and conditions with sender–receiver prioritization.	Needs biological replication, consistent annotation and adequate cells per relevant population.	Compare responder groups or disease phenotypes without treating cells as independent patients.
Tensor-cell2cell	Latent programs	Decomposes context-dependent communication into coordinated factors across samples, time or conditions.	Latent factors require careful stability assessment and biological interpretation.	Discover multi-interaction signatures that stratify heterogeneous patient contexts.
SpatialDM / COMMOT	Spatial evidence	Uses spatial co-expression or transport-aware modeling to localize candidate communication.	Resolution, spot composition, tissue geometry and ligand range shape interpretation.	Confirm that sender and receiver signals occupy a plausible tissue neighborhood.

Practical recommendation: use an interpretable primary method plus a consensus or orthogonal analysis when the decision is important. Tool agreement is supportive; disagreement should trigger review of interaction resources, expression filters, cell labels and sample composition—not automatic cherry-picking.

Patient-aware inference

Build a communication fingerprint per biological sample

A translational biomarker must exist at the level where the clinical outcome is defined: the patient. Pooling all cells in a response group can create attractive networks that cannot be assigned back to individual subjects.

Interaction activity

Ligand–receptor evidence in a defined sender–receiver pair

Receiver response

Pathway, transcription-factor or ligand-target activity

Cellular context

Relevant cell abundance, state frequency and tissue niche

Clinical association

Patient-level relationship with response or disease stratum

Assay feasibility

Detectable protein, transcript, cell phenotype or composite score

Design controls that preserve translation

Unit of inference: calculate sample-specific or replicate-aware communication summaries where the method permits.
Covariates: account for treatment time, site, batch, age, sex and other clinically relevant factors.
Paired designs: retain patient identity for pre/post-treatment samples rather than analyzing visits as unrelated.
Rare populations: distinguish true absence from insufficient sampling and report minimum-cell rules.
Validation split: prevent the same patients from informing feature selection and performance evaluation.

Biomarker qualification

A communication axis needs a translational passport

Prioritize candidates that travel well from discovery tissue to a practical assay. A biologically compelling interaction may still fail as a biomarker if it is unstable, unmeasurable or inseparable from sample composition.

Clinical contrast

Effect direction and magnitude are consistent at the patient level for the intended comparison, with uncertainty and prevalence reported.

Required

Mechanistic coherence

Sender ligand, receiver receptor and downstream response agree with one another and with the disease or treatment hypothesis.

Strengthens

Orthogonal support

Protein abundance, spatial proximity, perturbation, genetics or independent cohorts support at least one critical link in the chain.

Strengthens

Analytical robustness

Ranking survives reasonable annotation, filtering, normalization, prior-resource and method choices.

Required

Measurement route

The signal maps to an assayable analyte, cell phenotype or compact composite score in a clinically realistic specimen.

Required

Two translational routes

From interaction programs to clinical-use hypotheses

Patient stratification

Find communication programs that define biologically distinct subgroups before treatment or at diagnosis.

Derive sample-level interaction features.
Identify stable modules rather than isolated high scores.
Test association with molecular subtype, severity or outcome.
Reduce the module to an assay-compatible signature.
Validate subgroup reproducibility in an external cohort.

TRANSLATE

Efficacy prediction

Find pretreatment signals or early dynamic changes that distinguish benefit from resistance.

Define response labels and clinically appropriate time points.
Model baseline and on-treatment communication separately.
Link candidate axes to receiver-cell pharmacology.
Evaluate incremental value beyond standard biomarkers.
Lock thresholds before prospective or held-out testing.

Open-access evidence

Modern frameworks separate the task, context and evidence source

LIANA+ illustrates why cell-communication analysis is not a single score. Its framework integrates multiple ligand–receptor methods, multi-condition analysis, spatial relationships, extra-to-intracellular signaling and biological prior knowledge.

For translational work, this modularity matters: a candidate can be checked for consensus across methods, differential behavior across patient groups, downstream pathway consistency and spatial plausibility. These layers help turn a computational interaction into a testable biomarker hypothesis, but they remain inferential until supported by independent measurement or perturbation.

LIANA+ framework for single-cell, spatial and multi-condition cell communication analysis — **Figure 1, Dimitrov et al., 2024.** LIANA+ framework overview covering single-cell and spatial inference, multi-condition analysis, intracellular signaling and prior knowledge. Source: Nature Cell Biology 26, 1613–1622. Reproduced under CC BY 4.0.

Decision-ready outputs

What the analysis should deliver

Communication atlasSender–receiver maps, ranked ligand–receptor pairs, pathway summaries and directionality.

Patient feature matrixSample-level interaction and program scores ready for clinical association modeling.

Differential evidenceEffect sizes, uncertainty, covariate-adjusted comparisons and paired longitudinal results.

Biomarker shortlistCandidates ranked by clinical contrast, mechanism, robustness and assay feasibility.

Evidence ledgerAgreement across tools, resources, cohorts, spatial data and downstream-response analyses.

Validation blueprintRecommended specimen, assay, controls, thresholds and next experiment for each candidate.

Translational entry points

Related Services

Connect cell–cell communication findings with disease-mechanism interpretation, integrated functional evidence and dedicated single-cell RNA-seq analysis.

Mechanism resolutionPathway Analysis and Disease Mechanism Evidence integrationFunctional Genomics Data Interpretation Single-cell analysisSingle-Cell RNA-Seq Data Analysis

Start a Cell Communication Study

References

Selected scientific references

Dimitrov D, et al. LIANA+ provides an all-in-one framework for cell–cell communication inference. Nat Cell Biol. 2024;26:1613–1622. doi:10.1038/s41556-024-01469-w.
Dimitrov D, et al. Comparison of methods and resources for cell-cell communication inference from single-cell RNA-Seq data. Nat Commun. 2022;13:3224. doi:10.1038/s41467-022-30755-0.
Browaeys R, Saelens W, Saeys Y. NicheNet: modeling intercellular communication by linking ligands to target genes. Nat Methods. 2020;17:159–162. doi:10.1038/s41592-019-0667-5.
Jin S, et al. Inference and analysis of cell-cell communication using CellChat. Nat Commun. 2021;12:1088. doi:10.1038/s41467-021-21246-9.
Efremova M, et al. CellPhoneDB: inferring cell–cell communication from combined expression of multi-subunit ligand–receptor complexes. Nat Protoc. 2020;15:1484–1506. doi:10.1038/s41596-020-0292-x.
Armingol E, et al. Context-aware deconvolution of cell–cell communication with Tensor-cell2cell. Nat Commun. 2022;13:3665. doi:10.1038/s41467-022-31369-2.
Cang Z, et al. Screening cell–cell communication in spatial transcriptomics via collective optimal transport. Nat Methods. 2023;20:218–228. doi:10.1038/s41592-022-01728-4.
Li Z, et al. SpatialDM for rapid identification of spatially co-expressed ligand–receptor and revealing cell–cell communication patterns. Nat Commun. 2023;14:3995. doi:10.1038/s41467-023-39608-w.