The Two-Fold Bet: How Human Genetic Evidence Reshapes the Odds a Drug Target Succeeds

A drug target backed by human genetic evidence linking that target to the disease is roughly twice as likely to survive clinical development and reach approval as a target chosen without such evidence. Matthew Nelson and colleagues first quantified this in a 2015 Nature Genetics analysis, and a 2024 Nature study by Eric Minikel and colleagues refined the estimate, reporting that genetically supported target-indication pairs succeed about 2.6 times more often on the path from Phase I to launch. The signal is real and reproducible, but it is a shift in odds, not a guarantee, and its strength depends heavily on the kind of genetic evidence involved.

Why the target matters more than the molecule

Most drugs that enter clinical testing fail, and a large share fail because the biological hypothesis is wrong rather than the chemistry. The target simply does not drive the disease in humans the way the preclinical models suggested. This is the expensive lesson of modern drug development: efficacy failures in Phase II and Phase III are where fortunes and years disappear.

Human genetics offers a partial defense. If a naturally occurring variant in a gene shifts a person's risk of a disease, that gene is doing something causal in living human biology, not in a mouse or a dish. A drug that engages the same gene product is, in effect, testing a hypothesis that nature has already run. The premise of genetically anchored target selection is that these experiments of nature de-risk the central bet.

What Nelson 2015 actually showed

The 2015 paper by Nelson and colleagues in Nature Genetics compared drug mechanisms in the development pipeline against the catalog of gene-disease associations available at the time. Their headline finding, that selecting genetically supported targets could roughly double the success rate in clinical development, gave the field its durable rule of thumb. They also found that a meaningful fraction of existing drug mechanisms already had genetic support, suggesting the relationship was not a statistical accident but a structural feature of what makes targets work.

The phrasing that spread through the industry, that genetically supported targets are twice as likely to be approved, comes from this work. It became a slogan on pitch decks and in portfolio-strategy memos. Slogans lose their footnotes, and the footnotes here matter.

The 2019 correction that sharpened the claim

In 2019, a group publishing in PLOS Genetics asked directly whether targets with genetic support are twice as likely to be approved, and revised the estimate. Their answer was a qualified yes with an important split. Evidence from Mendelian disease genetics, the high-confidence variants cataloged in resources like OMIM, along with protein-altering coding variants, produced effects at or above the two-fold mark. Evidence drawn from genome-wide association studies (GWAS) produced a smaller and less consistent effect.

The reason is mechanistic. Most GWAS hits fall in noncoding regions of the genome, and translating a statistical association at a locus into confidence about a specific causal gene is genuinely hard. When a GWAS signal could be tied to a protein-altering change, its predictive value climbed toward the Mendelian level. The lesson was not that Nelson and colleagues were wrong, but that genetic support is not one thing. The confidence you can borrow from a variant depends on how clearly it points at a gene and how directly it perturbs the protein.

What Minikel 2024 refined

The 2024 Nature analysis by Minikel and colleagues is the most granular entry in this lineage. Working from tens of thousands of target-indication pairs drawn from pipeline data and integrating genetic evidence across Open Targets Genetics, the GWAS Catalog, OMIM, and large biobank cohorts including UK Biobank and FinnGen, they estimated a relative success ratio of about 2.6 from Phase I to approval for genetically supported pairs.

Two of their observations deserve emphasis. First, the source of evidence stratifies the benefit: Mendelian evidence from OMIM carried the highest relative success ratio, well above the GWAS-only signal, echoing the 2019 split. Second, and reassuringly for anyone building on GWAS, they reported that features once assumed to weaken a GWAS signal, such as smaller genetic effect size, common minor allele frequency, and the age of the discovery, did not meaningfully erode its predictive value. A modest, common-variant association can still carry real weight for target selection.

How to read the number without misusing it

The doubling figure is best understood as a prior, not a verdict. It reshapes the base rate of success across a portfolio; it does not promise that any single genetically supported program will work. A target can have impeccable genetic credentials and still fail on tolerability, pharmacology, or the gap between modifying a lifelong risk variant and reversing established disease with a drug given late.

Direction of effect is the subtle trap. Knowing that a gene matters is not the same as knowing which way to push it, and a variant that raises or lowers disease risk does not automatically tell you whether to inhibit or activate the corresponding protein. Genetic evidence points at a door; it does not always tell you whether to push or pull. The strength of the evidence also scales with its provenance, so a coding variant with a clear mechanism and a lone noncoding association at a crowded locus should not be treated as interchangeable inputs, even if both count as genetic support in a summary table.

Used well, this body of work justifies a deliberate tilt of a discovery portfolio toward targets that human biology has already implicated, while holding realistic expectations about how much any prior can carry. That is the honest version of the two-fold bet: better odds, earned by grounding the central hypothesis in human data, and not a shortcut around the hard work of proving a medicine safe and effective.

This article is educational and not medical advice.