When Nature Has Already Run the Experiment: Genetics and the Odds a Drug Target Works

A drug target that human genetics has already linked to a disease is roughly twice as likely to survive clinical testing and reach approval as a target chosen without that link. A 2015 Nature Genetics analysis by Nelson and colleagues first put a number on this idea, and a 2024 Nature study by Minikel and colleagues sharpened it, reporting that genetically supported target and indication pairs succeed about 2.6 times more often on the road from Phase I to launch. The effect is real and has held up across independent datasets, but it is a change in odds, not a promise, and how much it is worth depends heavily on what kind of genetic evidence is involved.

Key points#

• Most drugs that enter human testing fail, and many fail because the underlying biology was wrong, not the chemistry.
• Human genetic variants act as natural experiments: if a gene shifts disease risk in people, it is doing something real in human biology.
• Genetically supported targets reach approval about two to 2.6 times more often, a finding replicated from 2015 to 2024.
• Not all genetic support is equal. Rare, protein-changing (Mendelian) evidence carries more weight than a lone signal from a genome-wide association study.
• The doubling is a better starting bet, not a guarantee, and it does not tell you which direction to push a target.

The problem the number is trying to solve#

Bringing a medicine to market is mostly a story of failure. The large majority of programs that reach patients in early trials never make it to approval, and a stubborn share of those failures happen not because the molecule was poorly made but because the target was the wrong one. The biological hypothesis, that blocking or boosting a particular protein would change the course of a disease, simply did not hold in humans the way cells and animal models had suggested. These efficacy failures tend to arrive late, in Phase II and Phase III, after years of work and enormous cost.

That is the expensive gap human genetics tries to narrow. Every person carries millions of naturally occurring variants, and some of those variants nudge the risk of disease up or down. When a variant in a specific gene reliably changes who develops a condition, it is evidence that the gene is causally involved in living human biology. A drug aimed at the same gene product is, in a sense, betting on a hypothesis that nature has already been testing across whole populations for generations. Choosing targets this way does not remove risk, but it moves the central bet onto firmer ground.

Not all genetic evidence carries the same weight#

The single most useful thing to understand about this literature is that genetic support is not one uniform thing. The confidence you can borrow from a variant depends on how clearly it points to a particular gene and how directly it changes that gene's protein.

At the strong end sit rare, high-effect variants of the kind cataloged for Mendelian diseases, along with coding variants that alter a protein's sequence. These point almost unambiguously at a gene and a mechanism. At the weaker end sit many signals from genome-wide association studies (GWAS), where a statistical association marks a stretch of DNA rather than a named gene. Most of those signals fall in noncoding regions, and turning "something near here matters" into "this exact gene, acting this exact way" is genuinely hard. A single association at a crowded stretch of the genome and a clear protein-changing variant should not be treated as interchangeable inputs, even though a summary table might file both under "genetic support."

How the estimate was built and refined#

The 2015 work by Nelson and colleagues compared the mechanisms of drugs moving through the pipeline against the catalog of gene and disease associations known at the time. Their headline result, that picking genetically supported targets could roughly double the success rate in clinical development, became the field's durable rule of thumb. It also showed that a meaningful share of drugs already on the market had genetic support behind their targets, which suggested the pattern was structural rather than a fluke.

A 2019 PLOS Genetics study by King and colleagues asked the follow-up question directly: are genetically supported targets really twice as likely to be approved? The answer was a qualified yes with an important caveat. Evidence from Mendelian genetics and protein-altering coding variants produced effects at or above the two-fold mark, while GWAS-derived evidence produced a smaller and less consistent benefit. Notably, when a GWAS signal could be traced to a protein-changing variant, its predictive value climbed toward the Mendelian level. The takeaway was not that the original claim was wrong, but that its strength scales with the quality of the evidence behind it.

The 2024 Nature analysis by Minikel and colleagues is the most detailed entry in this lineage. Drawing on tens of thousands of target and indication pairs and pulling genetic evidence from resources such as Open Targets, the GWAS Catalog, OMIM, and large biobanks including UK Biobank and FinnGen, they estimated a relative success ratio near 2.6 from Phase I to approval for genetically supported pairs. Two of their findings stand out. First, the source of evidence again stratified the benefit, with Mendelian evidence carrying the highest success ratio. Second, and reassuringly, several features once assumed to weaken a GWAS signal, including a small effect size, a common variant, and an older discovery, did not meaningfully erode its usefulness. A modest, common-variant association can still earn its place in a target dossier.

Reading the number without abusing it#

The cleanest way to think about the doubling is as a prior, a sensible starting expectation, rather than a verdict on any one program. It reshapes the average odds across a portfolio of bets; it does not promise that a particular genetically supported drug will work. A target can have impeccable genetic credentials and still fail on safety, on the practicalities of drug delivery, or on the gap between nudging a lifelong risk variant and reversing an entrenched disease with a medicine given late.

Direction of effect is the subtle trap. Knowing that a gene matters is not the same as knowing which way to move it. Think of a thermostat: genetics can tell you the thermostat controls the room's temperature without telling you whether the room is too hot or too cold. A variant that changes disease risk does not automatically reveal whether the right drug should block the protein or stimulate it. That still has to be worked out.

Used well, this body of research supports a deliberate tilt toward targets that human biology has already implicated, paired with honest expectations about how far any starting bet can carry. That is the fair reading: better odds, earned by anchoring the core hypothesis in human data, and not a shortcut around the hard work of proving a medicine safe and effective.