Case Study

How Single-Cell RNA-Seq Improves Therapeutic Target Discovery

Inquiry

Technical Bulletin · Target Discovery

How Single-Cell RNA-Seq Improves Therapeutic Target Discovery

A decision-oriented framework for turning disease-associated cell states into ranked, testable and modality-aware target hypotheses.

Discuss Your Target Discovery Project

The translation gap

A marker gene is not yet a therapeutic target

Single-cell RNA sequencing can reveal which cell populations carry a disease program, which states expand or disappear, and which genes distinguish pathogenic cells from neighboring tissue. That resolution is valuable—but expression alone does not establish causality, tractability, safety or therapeutic direction.

The target-discovery task is therefore not to export the longest differential-expression table. It is to assemble a traceable argument: the candidate is connected to a disease-relevant cell state, the signal is reproducible across biological units, perturbing it is expected to change the phenotype, and an intervention can reach the target with an acceptable therapeutic window.

The strongest single-cell targets survive several independent questions—not merely one statistical ranking.

Where?Cell type, transient state, anatomical niche and disease context

Why?Regulatory, genetic, pathway and interaction evidence supporting mechanism

Can we act?Modality fit, selectivity, accessibility, assayability and safety

Core component 01

The candidate-target priority funnel

A funnel makes the attrition logic explicit. Each stage asks a different scientific question and records why a candidate advances, pauses or exits. The widths below are conceptual rather than fixed numerical cutoffs: projects should set thresholds around disease biology, tissue quality, cohort design and intended modality.

1 · Disease-linked single-cell signalcell identity · state abundance · differential program · niche

2 · Biological robustnessdonor-aware effects · covariates · external replication

3 · Mechanistic coherenceregulators · pathways · interactions · directionality

4 · Translational supporthuman genetics · disease specificity · safety context

5 · Actionabilitytractability · modality · accessibility · assays

Ranked shortlist
with evidence gaps and validation plan

Begin with a phenotype, not a gene.Define the pathogenic cell state and desired direction of change before ranking candidates.

Treat donors as the experimental units.Pseudoreplication can make abundant cells look more convincing than independent samples support.

Separate association from causation.Trajectory, ligand–receptor and regulatory-network outputs generate hypotheses; they do not prove a mechanism.

Keep rejected candidates auditable.An evidence ledger preserves exclusion reasons and supports future re-ranking when new data arrive.

Core component 02

A scoring matrix for comparable, reviewable decisions

The matrix converts heterogeneous evidence into a shared review surface for computational scientists, disease biologists, translational teams and chemists. The suggested weights are an illustrative starting point, not a universal industry standard. Reweighting should happen before candidate names are revealed whenever possible, reducing hindsight bias.

Dimension	Decision question	Evidence to inspect	Example weight
Disease–cell relevance	Is the candidate concentrated in the pathogenic cell state or niche that the therapeutic hypothesis seeks to modify?	Donor-aware differential expression, compositional change, state markers, spatial localization, case–control direction	18
Replication & robustness	Does the association persist across donors, cohorts, platforms and plausible analysis choices?	Pseudobulk models, covariate sensitivity, leave-one-donor-out checks, external datasets, effect-size concordance	15
Genetic or causal support	Is there orthogonal evidence connecting target perturbation or human variation to the disease phenotype?	GWAS fine-mapping, colocalization, rare variants, eQTL/pQTL, CRISPR or RNAi perturbation	15
Mechanistic coherence	Does the target occupy a plausible controlling position rather than merely report a downstream state?	Regulons, pathway position, temporal ordering, protein interactions, ligand–receptor context, prior biology	12
Tractability & modality fit	Can the target be modulated by an appropriate therapeutic modality and in the required direction?	Protein class, binding sites, surface accessibility, degradability, antibody or RNA suitability, precedent	12
Cell/tissue selectivity	Can efficacy-driving biology be separated from healthy-tissue expression?	Reference atlases, on-target/off-tissue expression, subcellular localization, disease-to-normal contrast	10
Safety liability	What on-target consequences are predicted in essential tissues or physiological processes?	Human loss-of-function data, essentiality screens, normal-tissue expression, phenotypes, paralog context	10
Experimental tractability	Can the hypothesis be challenged quickly in a disease-relevant system with a measurable endpoint?	Reagents, model availability, pharmacodynamic markers, rescue design, functional assays	8

0Unsupported, contradictory or not assessed

1Preliminary evidence or one context only

2Replicated, orthogonal or experimentally supported

Illustrative normalization: priority score = Σ(weight × evidence score / 2). Report the total together with confidence, missingness and any veto—not as a standalone truth. Two candidates with identical totals may require different next experiments.

Do not average away fatal flaws

Evidence vetoes and penalties

A weighted score can conceal a decisive weakness. Establish vetoes before ranking and distinguish an evidence gap, which may be experimentally resolvable, from evidence against the hypothesis.

Single-donor or batch-confounded signal: apparent specificity disappears under donor-aware modeling or technical covariate checks.
Marker-only logic: expression labels the state but provides no reason to expect that modulation will alter disease biology.
Direction mismatch: the intended agonism or inhibition conflicts with genetics, perturbation results or disease-stage biology.
Ubiquitous essential function: disease-cell relevance is outweighed by dependence in critical normal tissues.
Modality–location mismatch: the proposed modality cannot reach the target compartment or the relevant cells.
Model mismatch: validation relies on a species, cell line or stimulation context that does not reproduce the human disease state.

From rank to evidence

A validation ladder that retires uncertainty in stages

Prioritization should dictate the next most informative experiment, not simply produce a top-ten list. The aim is to remove the most consequential uncertainty at the lowest reasonable cost while preserving disease relevance.

Computational triangulationRe-test cell-state association with donor-aware statistics; examine external cohorts, genetics, regulatory context and healthy-tissue atlases.

Orthogonal localizationConfirm RNA and protein in the predicted cells or niche using targeted assays, imaging, flow cytometry or proteomic evidence.

Perturbation with specificity controlsUse multiple reagents, dose response and rescue where feasible; verify target engagement and separate general toxicity from mechanism.

Disease-relevant functional phenotypeMeasure an endpoint tied to the therapeutic hypothesis in primary cells, co-culture, organoid or other context-appropriate systems.

Translational confirmationEvaluate exposure, pharmacodynamic markers, efficacy and safety in a model that preserves the relevant biology; define criteria for progression.

Open-access evidence

Prioritization is useful only when followed by functional challenge

Sokol and colleagues applied a structured target-assessment framework to a set of single-cell-derived endothelial tip-cell markers. Their study illustrates the essential distinction between a ranked marker list and a testable target program: candidates were filtered using target–disease linkage, target-related safety, strategic considerations and technical feasibility, then advanced to in vitro and in vivo validation.

The figure below is particularly relevant to decision-oriented target discovery because it shows attrition from 50 genes to six prioritized candidates and connects selection directly to functional assays. The authors reported that not every high-ranking marker behaved as predicted, reinforcing the need to keep computational priority and functional truth separate.

Study design and staged prioritization of single-cell-derived endothelial target genes — **Excerpt from Figure 1a–b, Sokol et al., 2023.** Panel a connects prioritized markers with in vitro and in vivo functional validation; panel b shows staged filtering by target–disease linkage, target-related safety, strategic considerations and technical feasibility. Panels c–d were omitted because they describe case-specific abundance and expression results rather than the general prioritization framework discussed here. Source: Communications Biology 6, 648. Article distributed under CC BY 4.0; excerpted without altering the retained panels. BioRender-created elements remain subject to their stated credit.

Decision-ready deliverable

What a target dossier should contain

A useful output makes every claim traceable to data and every uncertainty actionable. Instead of a static gene list, package each shortlisted candidate as a compact target dossier.

Cell-state evidence

Annotation confidence, affected population, donor-level effect, disease direction, niche and cohort context.

Mechanistic hypothesis

Expected role, intervention direction, regulatory or interaction evidence, and alternative interpretations.

Translational context

Human genetic support, normal-tissue expression, tractability, precedent, safety considerations and modality fit.

Evidence ledger

Supporting, neutral and contradictory findings with source, analysis version and confidence level.

Scorecard & sensitivity

Dimension-level scores, weights, missingness, veto status and changes under alternative weighting assumptions.

Next-best experiment

Specific hypothesis, model, perturbation, readout, controls and predefined go/no-go interpretation.

Project initiation

Minimum inputs for a defensible prioritization program

The strongest project brief aligns the disease question, dataset and decision criteria before analysis begins.

Target product hypothesisDisease, patient segment, desired biological effect and intended intervention direction.

Biological comparisonCase/control, responder/non-responder, stage, tissue compartment and critical covariates.

Sample metadataDonor IDs, batches, treatments, collection sites, quality metrics and paired observations.

Analysis-ready dataRaw counts or validated object, reference build, feature annotation and upstream processing history.

External evidence sourcesRelevant cohorts, genetics, perturbation screens, atlases and internal experimental findings.

Decision rulesWeights, vetoes, acceptable uncertainty, shortlist size and the experiment that follows ranking.

Start a Target Prioritization Discussion

Related capabilities

Services Supporting Target Discovery

References

Selected scientific references

Sokol L, et al. Prioritization and functional validation of target genes from single-cell transcriptomics studies. Commun Biol. 2023;6:648. doi:10.1038/s42003-023-05006-7.
McDonagh EM, et al. Human Genetics and Genomics for Drug Target Identification and Prioritization: Open Targets' Perspective. Annu Rev Biomed Data Sci. 2024;7:59–81. doi:10.1146/annurev-biodatasci-102523-103838.
Ochoa D, et al. The next-generation Open Targets Platform: reimagined, redesigned, rebuilt. Nucleic Acids Res. 2023;51(D1):D1353–D1359. doi:10.1093/nar/gkac1046.
Mountjoy E, et al. An open approach to systematically prioritize causal variants and genes at all published human GWAS trait-associated loci. Nat Genet. 2021;53:1527–1533. doi:10.1038/s41588-021-00945-5.
Replogle JM, et al. Mapping information-rich genotype–phenotype landscapes with genome-scale Perturb-seq. Cell. 2022;185:2559–2575.e28. doi:10.1016/j.cell.2022.05.013.
Dixit A, et al. Perturb-Seq: dissecting molecular circuits with scalable single-cell RNA profiling of pooled genetic screens. Cell. 2016;167:1853–1866.e17. doi:10.1016/j.cell.2016.11.038.
Finan C, et al. The druggable genome and support for target identification and validation in drug development. Sci Transl Med. 2017;9:eaag1166. doi:10.1126/scitranslmed.aag1166.
Plenge RM, Scolnick EM, Altshuler D. Validating therapeutic targets through human genetics. Nat Rev Drug Discov. 2013;12:581–594. doi:10.1038/nrd4051.