From GWAS to Drug Targets: Integrating Genetic Evidence, QTL Mapping, and Colocalization

A genome-wide association study identifies a statistically significant association between a genomic region and a disease. That association is a starting point, not an answer. The variant is in a non-coding region. Several genes lie nearby. The biological mechanism is unknown. And the question that actually matters, which protein should a drug target, and why will modulating it help patients, remains entirely open.

The gap between a GWAS signal and a validated drug target is where most of the hard work of human genetics-based drug discovery sits. This article explains what that gap is, why it exists, and what the integrated toolkit of fine-mapping, QTL analysis, colocalization, and Mendelian randomization does to close it.

Why GWAS Is a Beginning, Not an End

GWAS has transformed our understanding of the genetic architecture of complex disease. The NHGRI-EBI GWAS Catalog contains over 500,000 variant-trait associations from more than 6,000 publications as of 2025, and the pace of discovery continues to accelerate as biobank cohorts grow. The method works by scanning millions of genetic variants across the genome and identifying those statistically associated with a phenotype in a large population.

The central challenge is that more than 90% of GWAS-associated variants map to non-coding regions of the genome. They are not in protein-coding exons where a direct functional consequence is obvious. They fall in introns, intergenic regions, enhancers, and promoters, where their effect on biology is indirect, cell-type-specific, context-dependent, and not immediately interpretable. A systematic literature review identified only 309 experimentally validated non-coding GWAS variants across 130 human disease traits, despite tens of thousands of reported associations. The bottleneck is not statistical power. It is functional interpretation.

This is the post-GWAS problem: a locus associated with disease contains many variants in high linkage disequilibrium with each other, any one of which could be causal, potentially affecting any nearby gene, in any cell type relevant to the disease, through any of several regulatory mechanisms. Getting from that locus to a gene, and from a gene to a protein target, requires a layered analytical strategy.

Step 1: Fine-Mapping - From Region to Credible Variants

The first task after identifying a GWAS association is fine-mapping: reducing the list of potentially causal variants within the associated region from dozens or hundreds of linked variants to a much smaller credible set with the highest posterior probability of containing the causal variant.

Fine-mapping methods use the linkage disequilibrium structure of the region and the association statistics to estimate which variants are most likely causal. Methods such as FINEMAP, SuSiE (Sum of Single Effects), and POLYFUN use different statistical approaches to compute posterior inclusion probabilities for each variant and generate credible sets, analogous to confidence intervals in the variant space rather than in effect size space. A well-powered fine-mapping analysis might narrow a region of 300 correlated variants to a credible set of 3 to 10, of which one or a small number are expected to be causal.

Fine-mapping is a critical prerequisite for all downstream analysis. Its importance as a key step for subsequent colocalization and gene prioritization has been consistently demonstrated. The choice of fine-mapping method affects which variants are prioritized, which in turn affects which genes and mechanisms are implicated downstream, which is why fine-mapping and colocalization results from the same GWAS can vary substantially depending on which tools are applied.

Step 2: QTL Mapping - Connecting Variants to Molecular Phenotypes

Once a credible set of variants is identified, the next question is: what do these variants do biologically? For non-coding variants, the most common answer involves regulatory effects on gene expression, and the primary tool for identifying these effects is quantitative trait locus (QTL) mapping.

An expression QTL (eQTL) is a genetic variant associated with the expression level of a gene, typically measured as mRNA abundance in a specific tissue. A protein QTL (pQTL) is similarly associated with protein abundance. These molecular QTLs provide a mechanistic link: if a GWAS-associated variant is also an eQTL for a nearby gene in a relevant tissue, it suggests the variant influences disease risk by modulating the expression of that gene in that context.

Large-scale eQTL resources, most prominently the GTEx (Genotype-Tissue Expression) project, have mapped cis-eQTLs across 54 human tissues and more than 900 individuals, providing a comprehensive reference for which genetic variants regulate gene expression in which tissues. Disease-relevant tissues can be queried: variants associated with an inflammatory disease can be examined for eQTL effects in relevant immune cell types, rather than in tissues unlikely to be involved in the disease.

Integrating molecular QTL data with GWAS signals substantially improves the identification of causal genes. A study combining variant annotation, activity-by-contact maps, Mendelian randomization, and colocalization with molecular QTL data across 4,611 disease GWAS from FinnGen, the Estonian Biobank, and UK Biobank found that genes supported by multi-omics evidence displayed higher enrichment in approved therapeutic targets compared to the nearest gene mapping approach alone, with a risk ratio of 1.75 versus 1.56. The analysis also found that predicted directionality from Mendelian randomization was consistent with matched drug mechanism of action in more than 85% of approved drugs, providing a direct link between QTL-derived causal inference and pharmacological reality.

Step 3: Colocalization - Distinguishing Shared Signals From Coincidence

Finding that a GWAS-associated variant is also an eQTL in a relevant tissue is suggestive, but not sufficient, evidence for a causal relationship. The association and the eQTL could overlap by chance, each driven by a different underlying variant in the same region that happens to be in linkage disequilibrium with the other. Colocalization analysis is designed to distinguish genuine shared signals from this coincidental overlap.

Colocalization methods test whether a GWAS association and a QTL association in a specific tissue are driven by the same underlying causal variant, rather than by different variants in the same genomic region. The most widely used method, COLOC, computes posterior probabilities for five hypotheses: no association in either dataset, an association specific to the GWAS, an association specific to the QTL, two distinct causal variants in the same region, and a single shared causal variant driving both associations. The fifth hypothesis, shared causal variant, is the target. A high posterior probability for hypothesis five (typically above 0.8) is taken as evidence that the variant affects both gene expression and disease risk through the same molecular mechanism.

This distinction matters because non-shared overlap is common. Two signals can overlap in the same region without being driven by the same variant, and treating that overlap as mechanistic evidence when it is coincidental leads to incorrect gene prioritization. Fine-mapping and colocalization approaches are specifically designed to identify causal genetic variants with a biological effect on the trait and their functional effects, and their joint application substantially reduces the rate of false positive gene nominations relative to simpler proximity-based approaches.

Colocalization has real limitations. It is sensitive to the statistical model used, the reference population for linkage disequilibrium estimation, and the prior probabilities assigned to each hypothesis. Systematic benchmarking has shown that the choice of colocalization method significantly affects which genes are prioritized as candidates. Performance also varies with the extent of co-regulation of neighboring genes: extensive co-regulation in a locus can complicate eQTL-based gene prioritization, and recent benchmarking found that assigning fine-mapped protein QTLs to their closest protein-coding gene actually outperformed all tested colocalization methods on precision and recall in one large analysis. The field consensus is that prioritizing novel targets requires triangulation of evidence from multiple sources, rather than reliance on any single method.

Step 4: Mendelian Randomization - Estimating Causal Effects

Colocalization identifies which gene is likely regulated by the disease-associated variant. Mendelian randomization (MR) asks the next question: does changing the level of that gene's product causally affect disease risk?

MR uses genetic variants as instrumental variables to estimate the causal effect of an exposure (such as the expression level or activity of a protein) on an outcome (disease risk). Because genetic variants are assigned at conception and are not influenced by environmental confounding, they provide a natural experiment that is analogous, in principle, to random assignment in a clinical trial. If variants that increase the expression of gene X also increase disease risk, and those variants do not plausibly affect disease through any pathway other than gene X, MR provides evidence that gene X causally influences disease.

The application of MR to drug target validation, called drug-target MR, uses variants in or near the gene encoding a drug's target as instruments for the target's activity. The HMGCR gene, which encodes HMG-CoA reductase, the molecular target of statins, provides the canonical illustration. Variants in HMGCR associated with lower LDL cholesterol predict reductions in coronary artery disease risk with effect sizes that closely mirror those observed in statin clinical trials, providing genetic confirmation that the therapeutic effect of statins operates through their intended target. The same approach has been applied to PCSK9 to predict the cardiovascular effects of PCSK9 inhibitors before those drugs completed large-scale trials, and similar analyses using HMGCR and PCSK9 variants have now been used to investigate potential on-target effects on other diseases including Alzheimer's risk and primary sclerosing cholangitis.

The value of drug-target MR is that it provides evidence about what will happen if a drug modulates a specific target in humans, using human genetic data, before expensive clinical trials are conducted. It does not guarantee clinical success, and MR has its own assumptions and limitations, including the requirement that genetic instruments affect the disease only through the pathway of interest and not through pleiotropic effects. But when drug-target MR results are consistent with clinical trial outcomes, it validates both the approach and the target.

Why This Pipeline Predicts Clinical Success

The mechanistic rationale for why genetic evidence predicts clinical success is straightforward: a drug working through a target whose disruption demonstrably causes or prevents disease in humans is more likely to produce the desired clinical effect than a drug working through a target whose connection to disease is based on preclinical models or correlative biology alone.

The quantitative evidence supports this. An analysis by Minikel and colleagues published in Nature in 2024 estimated that drug mechanisms with genetic support are 2.6 times more likely to succeed in clinical development than those without, updating the prior estimate of a two-fold improvement with a larger evidence base and more refined analysis. The effect holds across therapy areas and development phases, improves when the causal gene is identified with greater confidence, and is largely unaffected by genetic effect size or variant allele frequency. These results indicate we are far from reaching peak genetic insights to aid the discovery of targets for more effective drugs.

The improvement in success is greatest when the causal gene can be identified unambiguously, as in Mendelian disease where a single gene's disruption causes disease, or when a GWAS association is linked to a coding variant with a direct effect on protein function. GWAS associations in non-coding regions, where causal gene identification requires the full fine-mapping and colocalization pipeline described above, also improve success rates, but to a lesser extent, reflecting the remaining uncertainty in causal gene assignment. The implication is that investments in improving fine-mapping precision, expanding QTL resources across more cell types and disease-relevant contexts, and developing more accurate colocalization methods all have direct downstream value for drug development.

The Literature Synthesis Bottleneck

The technical pipeline described above generates candidate genes and directional hypotheses: gene X, expressed in tissue Y, causally affects disease Z through mechanism W. But connecting that output to the existing published biology of those genes in disease-relevant contexts, to prior drug discovery efforts targeting the same pathway, and to the mechanistic literature that explains how the gene product might be modulated pharmacologically, requires synthesis across a scientific literature that has grown too large for manual curation to handle comprehensively.

A post-GWAS framework that produces a prioritized gene list but cannot rapidly connect those genes to the mechanistic disease context leaves a gap between the statistical output and the drug discovery decision. Whether gene X is a viable drug target depends not only on the genetic evidence but on its protein structure, its druggability, its expression pattern across healthy and disease tissue, prior attempts to target related biology, and what the mechanistic literature reveals about its role in disease pathogenesis. Assembling that evidence systematically, across multiple candidate genes simultaneously, in a way that is anchored to the specific disease and cell type context, is where context-aware literature synthesis tools can extend the reach of the post-GWAS pipeline.

The Practical Picture

The pipeline from GWAS to drug target is not a linear algorithm that produces a definitive answer. It is an evidence integration framework that progressively increases confidence in a gene's causal role while generating testable hypotheses for experimental validation. Fine-mapping reduces the variant space. eQTL and pQTL data connect variants to genes and tissues. Colocalization distinguishes shared signals from coincidental overlap. Mendelian randomization estimates the causal direction and magnitude of the gene's effect on disease. And the output of all of that feeds into target selection, where it is combined with druggability assessment, competitive intelligence, and mechanistic biology before a therapeutic program begins.

None of these steps alone is sufficient. Prioritizing novel targets requires triangulation of evidence from multiple sources, and the value of the pipeline increases substantially as each layer of evidence is added. The 2.6-fold improvement in clinical success associated with genetic evidence is the aggregate result of working through this pipeline systematically, and the gap between that 2.6-fold improvement and a much larger one represents the work that remains in connecting GWAS signals to causal biology with the precision that drug development requires.

---

References

1. Minikel EV, Painter JL, Dong CC, Nelson MR. Refining the impact of genetic evidence on clinical success. Nature. 2024;629:624. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11096124/

2. King EA, Davis JW, Degner JF. Are drug targets with genetic support twice as likely to be approved? PLOS Genetics. 2019;15(12):e1008489. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6907751/

3. Tambets R, Kolde A, Kolberg P, Love MI, Alasoo K. Extensive co-regulation of neighboring genes complicates the use of eQTLs in target gene prioritization. HGG Advances. 2024;5(4):100348. https://doaj.org/article/563ae9b155174a34a141d7ee3460a524

4. Pérez-Granado J, Piñero J, Furlong LI. Benchmarking post-GWAS analysis tools in major depression: Challenges and implications. Frontiers in Genetics. 2022;13:1006903. https://www.frontiersin.org/journals/genetics/articles/10.3389/fgene.2022.1006903

5. Lopes RR et al. Leveraging large-scale multi-omics evidences to identify therapeutic targets from genome-wide association studies. BMC Genomics. 2024. https://doaj.org/article/25178c318afb4bd683f1e5f134428e3c

6. Sinnott-Armstrong N et al. The landscape of GWAS validation: systematic review identifying 309 validated non-coding variants across 130 human diseases. BMC Medical Genomics. 2022. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8973751/

7. Wurtz P et al. Characterising metabolomic signatures of lipid-modifying therapies through drug target Mendelian randomisation. PLOS Biology. 2022;20(2):e3001547. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8906647/

8. Swerdlow DI. Validation of therapeutic targets in cardiometabolic disease using Mendelian randomisation. UCL PhD Thesis. 2012. https://discovery.ucl.ac.uk/id/eprint/1355962/

9. Ramos-López AI et al. From GWAS signals to molecular mechanisms: explainable AI for causal gene prioritization and biomolecular target interpretation. Biomolecules. 2025. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13406849/