Drug Target Validation Using CRISPR Screens, Perturb-seq, and Functional Genomics

More than 90% of compounds that enter clinical trials fail to demonstrate sufficient safety and efficacy to gain regulatory approval. The most common reason is not toxicity. It is lack of efficacy, and the root cause most frequently traced to efficacy failure is an incorrect assumption at target selection: the target chosen did not play the causal role in the disease that was assumed when the program began.

This failure mode is old, well-documented, and partially addressable. Functional genomics, and specifically CRISPR-based screening combined with single-cell transcriptomics through approaches like Perturb-seq, has emerged as the most direct method available to establish whether a gene plays a causal role in a disease-relevant phenotype, in human cells, before committing to a development program. Understanding what these tools do, how they connect to clinical decision-making, and where their own limitations sit is increasingly important for anyone working at the intersection of genomics and drug discovery.

Why Target Validation Is the Rate-Limiting Step

The analysis of pharmaceutical industry data indicates that the major reason for drug candidates failing in late-stage clinical development is lack of efficacy, with a high proportion due to erroneous hypotheses about target-to-disease linkage. Drug candidates with adequate pharmacokinetics and safety margins still commonly fail because the wrong target was selected as the focus of the therapeutic program.

The conventional routes to target identification, comparative genomics, transcriptomics, and proteomics, are powerful for detecting associations between molecular variables and disease states, but they primarily establish correlation rather than causation, and they are often confounded by passenger alterations and differential expression with limited functional relevance. Functional genomics can overcome these limitations by directly annotating gene functions, establishing causal links between genes and diseases, rather than inferring them from expression correlations.

The practical implication is well-established. Drug targets supported by human genetic evidence are more than twice as likely to succeed in clinical development as those without such evidence. Functional genomics tools, when layered on top of genetic evidence, extend that causal chain from population-level association to cellular mechanism.

What CRISPR Screens Actually Do

A CRISPR screen perturbs one gene at a time across a population of cells and measures the effect of each perturbation on a phenotype of interest. The most common format is a pooled screen: a library of guide RNAs, each targeting a different gene, is introduced into a cell population, the cells are cultured under the condition of interest (disease-relevant stress, drug treatment, infection, and so on), and the relative abundance of each guide RNA in the surviving population is measured by sequencing. Guides that are depleted indicate genes whose loss impairs survival or the phenotype of interest. Guides that are enriched indicate genes whose loss confers resistance or benefit.

CRISPR screening has become a central technology in functional genomics, enabling genome-scale interrogation via pooled perturbations. Three key CRISPR modalities are in regular use. CRISPRko uses Cas9 to knock out genes by inducing double-strand breaks. CRISPRi uses a catalytically dead Cas9 fused to a transcriptional repressor to reduce gene expression without altering the DNA sequence. CRISPRa uses a dead Cas9 fused to an activator to increase gene expression. Each modality answers a slightly different question: knockout tests gene essentiality, CRISPRi tests the effect of reduced expression, and CRISPRa tests the effect of increased expression, which can mimic gain-of-function disease states.

The readout in a conventional pooled screen is a single number per guide RNA: its relative abundance. This is powerful for phenotypes that can be captured by cell survival or proliferation, but it collapses the complexity of what happens inside each cell into a binary signal.

Perturb-seq: Adding Transcriptomic Resolution

Perturb-seq combines CRISPR perturbation with single-cell RNA sequencing to measure not just whether a cell survived after a gene was perturbed, but what the full transcriptional state of that cell looks like. Each cell in the experiment carries a guide RNA that identifies which gene was perturbed, and single-cell sequencing reads out the transcriptome of that individual cell simultaneously. The result is a dataset in which each gene perturbation is linked to the transcriptional signature it produces.

This multimodal readout changes what can be learned. In a conventional pooled screen, a hit is a gene whose loss changes the competitive fitness of cells. In a Perturb-seq screen, a hit is a gene whose loss produces a specific transcriptional program, which can be cross-referenced against disease-associated gene expression signatures, known pathway activation states, and the transcriptional consequences of drug treatment. Perturb-seq enables the systematic identification of the targets of signaling regulators across diverse biological contexts, producing pathway signatures that can be used to infer changes in signaling pathway activation in in-vivo and in-situ samples.

The pharmaceutical application is direct. If the transcriptional consequence of perturbing gene X closely resembles the transcriptional signature of the disease of interest, gene X is a candidate causal driver. If a drug is known to produce a specific transcriptional state, Perturb-seq data can be used to identify which gene perturbations phenocopy that state, providing a hypothesis about the drug's mechanism of action. A Bayesian regression approach that explains drug treatment effects as combinations of perturbation effects has been validated on published single-cell perturbation and drug screen datasets, achieving a median sensitivity of 92% and median precision of 99% in classifying cellular perturbations and correctly identifying BCR as the highest-probability target of BCR-ABL inhibitors.

The critical practical limitation of Perturb-seq has historically been scale. Profiling every cell in a genome-wide perturbation experiment at single-cell resolution is expensive and computationally demanding. Recent technical advances are addressing this constraint directly. VIPerturb-seq, developed using probe-based detection workflows compatible with fixed-cell single-cell RNA-seq, demonstrated an unbiased genome-wide screen profiling 880,000 cells performed in a single afternoon, making genome-wide single-cell screening practically accessible to biopharma target discovery programs. Xaira Therapeutics has generated X-Atlas/Pisces, described as the largest genome-wide CRISPRi Perturb-seq compendium to date, comprising 25.6 million perturbed single-cell transcriptomes across 16 biologically diverse contexts including primary cell types and iPSC-derived lineages.

The Perturbomics Framework

The term perturbomics describes the systematic analysis of phenotypic changes resulting from gene function modulation, of which CRISPR-based screening is the primary implementation. CRISPR screens have become the method of choice for perturbomics studies, enabling the identification of target genes whose modulation may hold therapeutic potential for diseases such as cancer, cardiovascular disorders, and neurodegeneration, contributing to the development of targeted drug therapies and the design of gene and cell therapies for regenerative medicine.

The perturbomics workflow involves several stages that are worth understanding separately. The initial screen, whether pooled or arrayed, identifies a set of candidate genes associated with the phenotype of interest. This is a discovery step, not a validation step. Positive hits from the initial screen require follow-up experiments, such as individual gene knockouts or orthogonal validation with independent perturbation modalities, to confirm their functional relevance to the phenotype of interest before being treated as validated targets.

Arrayed screens, in which each gene perturbation is performed in a separate well and phenotype is measured directly by imaging or other high-content readouts, complement pooled screens by enabling rich per-perturbation phenotypic readouts including morphology, protein localisation, and multi-marker measurements. The long-term success of target discovery is currently based on four key pillar areas: the creation and implementation of successful array screening modalities, the use of translatable and physiologically relevant disease models, the integration of complex assay readouts, and the establishment of bioinformatic pipelines for quantitative analysis of phenotypic readouts.

Connecting Screens to Clinical Evidence

Functional genomic screens establish that a gene matters for a phenotype in a cell-based model. To connect that observation to clinical relevance, two additional layers of evidence are typically required.

The first is genetic evidence in human populations. A gene whose perturbation in a CRISPR screen produces a disease-relevant phenotype, and whose natural variants are associated with that disease in GWAS or sequencing studies, represents a convergence of experimental and population-level evidence that substantially strengthens the case for causal involvement. Utilization of patient genetics as target validation is yielding targets and mechanisms with an estimated two-fold increase in success in the clinic, consistent with the broader estimate that genetic support doubles clinical success rates.

AstraZeneca has built a target discovery platform using genome-wide CRISPRn and CRISPRa libraries specifically designed to deliver new target opportunities with an increased chance of success in the clinical development programme, integrating genomic data and CRISPR gene editing for identification and validation of new drug targets. Genetic linkage to disease and model translatability are identified as critical features associated with reduced clinical efficacy failure in their framework.

The second is model translatability. A screen run in an immortalised cancer cell line establishes gene function in that specific context. Reproducing the finding in primary human cells, in patient-derived organoids, or in iPSC-derived disease-relevant cell types substantially increases confidence that the target biology is relevant to the actual human disease rather than an artefact of the model. Single-cell CRISPR technologies are designed to enable more physiologically relevant, single-cell-level analyses precisely to address this translational gap.

The Limitations That Remain

CRISPR-based functional genomics addresses the driver-passenger problem more directly than any previous technology, but it does not resolve it completely.

A gene essential for the survival of a cancer cell line in a pooled screen may be essential because of a specific genetic background in that line rather than because of the disease mechanism being targeted. The same gene may be dispensable in a more representative model. This is the context-specificity problem: the target biology identified in a screen is always conditional on the model in which the screen was run.

Guide RNA off-target effects introduce noise into screen data, though improvements in guide RNA design and the availability of CRISPRi as an alternative to knockout have substantially reduced but not eliminated this problem. Many drugs, including some advanced into clinical trials, act through targets that differ from their reported mechanism, complicating safety assessment and limiting therapeutic efficiency.

And CRISPR screens, however large, remain artificial perturbations. Completely abolishing a gene's function, as CRISPRko does, is not what most drugs do. Most drugs partially inhibit a target, in a specific conformation, with dose-dependent effects that vary across cell types. A gene that is essential for cell survival when knocked out may be an excellent drug target, a poor one, or something in between, depending on what happens when its function is only partially reduced by a therapeutic.

Where the Field Is Moving

The trajectory is toward more physiologically relevant models, richer multimodal readouts, and systematic integration with genetic and clinical evidence. Combining genome-scale CRISPR screening with phenotype-based enrichment and multimodal single-cell profiling, as in recent Perturb-ME workflows, converts a genome-wide enrichment screen from a hit-discovery assay into a mechanism discovery tool, enriching for the cellular states most relevant to the disease biology before applying the expensive single-cell readout.

The connection to causal biology modeling is explicit in the most ambitious current implementations. Causal models of cellular systems hold the promise to empower broad biological discovery, including the systematic identification of novel targets for drug discovery. Large-scale Perturb-seq compendiums, when combined with foundation models trained on perturbation data, are beginning to enable in silico prediction of perturbation effects that can guide experimental prioritization rather than requiring every perturbation to be tested empirically.

The practical endpoint, a comprehensive, cell-type-specific map of how perturbing each gene in the human genome changes cellular state, is now a technically tractable goal rather than a conceptual one. How that map connects to clinical decision-making, and how the field validates that connection before committing programs, remains the central question for functional genomics in drug discovery.

---

References

1. Park B-S et al. Perturbomics: CRISPR-Cas screening-based functional genomics approach for drug target discovery. Experimental and Molecular Medicine. 2025;57:1443-1454. https://pmc.ncbi.nlm.nih.gov/articles/PMC12322284/

2. Chen K et al. CRISPR screening redefines therapeutic target identification and drug discovery with precision and scalability. Journal of Pharmaceutical Analysis. 2025. https://pmc.ncbi.nlm.nih.gov/articles/PMC12925547/

3. Li C et al. High-content CRISPR screening: methods and applications. Advanced Science. 2025. https://pmc.ncbi.nlm.nih.gov/articles/PMC13551093/

4. Single-cell CRISPR technologies review. Blood Science. 2025. https://pmc.ncbi.nlm.nih.gov/articles/PMC12680425/

5. Jiang L et al. Systematic reconstruction of molecular pathway signatures using scalable single-cell perturbation screens. bioRxiv. 2024. https://doi.org/10.1101/2024.01.29.576933

6. VIPerturb-seq: genome-wide CRISPR screens at single-cell resolution. 10x Genomics / New York Genome Center. 2026. https://www.pharmtech.com/view/viperturb-seq-genome-wide-crispr-screens-at-single-cell-resolution

7. Chu C, Wang B. Towards causal biology modeling: industrializing high-throughput Perturb-seq at genome scale. Xaira Therapeutics / Broad Institute. 2026. https://www.broadinstitute.org/node/5561016

8. Columbia University. Method to identifying drug targets from single-cell perturbation screens. Tech Ventures. https://inventions.techventures.columbia.edu/technologies/method-to-identifying-drug--CU25189

9. Gianni D, Escudero-Ibarz L. Functional genomics: arrayed CRISPR KO screens. Expert Opinion on Drug Discovery. 2022;9:127-136. https://knowledge.lonza.com/citation?id=6761

10. Plenge RM et al. Validating therapeutic targets through human genetics. Nature Reviews Drug Discovery. 2013. https://www.nature.com/articles/nrd4051.pdf

11. AstraZeneca / Medicines Discovery Catapult. Target discovery at AstraZeneca: discovery, identification, and novel targets. https://md.catapult.org.uk/blogs/target-discovery-at-astrazeneca-discovery-identification-and-novel-targets/