SP-03 · Is This Knot Real?

The target has a name now. Stage 2 ended with a nomination — a bet, assembled from genetics and pathway maps and a story that hangs together, that this molecule is the knot in the disease. A protein doing the wrong thing, or too much of the right thing, somewhere it shouldn’t. The case is good enough to write down. It is not good enough to spend forty million dollars on.

So before anyone spends that money, a different question has to be asked, and it is not a comfortable one. The nomination says the target is associated with the disease — it shows up where the disease is, it moves when the disease moves. But association is cheap. A fire truck is reliably present at fires. The question that separates a real drug program from an expensive education is narrower and harder: if you reach into a living human body and shut this one protein down, does the disease actually get better — or do you just learn, slowly and at great cost, that you were looking at the fire truck?

That question is the whole of this stage. It has a name — target validation — and it is best understood not as a technique but as a discipline: the discipline of building an honest, causal case that hitting your target will change the disease in a person, before you commit the money that makes turning back almost impossible. It is also, quietly, the discipline of resisting your own excitement. By the time a target reaches validation, a team is already in love with it. The work of this stage is to fall out of love on purpose, and see whether the evidence brings you back.

The Fallacy You Have to Disarm First

Picture the moment validation usually turns. A team has a knockout mouse — an animal engineered so the gene for the target is switched off — and in that mouse, the disease doesn’t appear. Take away the protein, and the pathology is gone. Better still, the result reproduces: three separate labs run the experiment and three separate times the mouse comes back healthy. There is a room, and a slide deck, and a number on the slide — tens of millions of dollars — and the result on the screen is as clean as biology ever gets. The case for greenlighting the spend feels overwhelming.

It feels overwhelming because of a sentence almost no one says out loud but nearly everyone half-believes: if the mouse works, the human will work. That sentence is the most expensive idea in this stage, and it is wrong. Not wrong because mice are useless — they are not — but wrong about what the mouse result is. The clean, reproducing animal phenotype is the most seductive evidence in all of drug discovery, and for exactly that reason the most treacherous.

Here is the correction, and it is worth holding onto for the rest of the curriculum: animal evidence filters; it does not predict. A good mouse model is a screen that catches failures — if your target’s gene is deleted and the disease gets worse, or the mouse dies, you have learned something fast and cheap, and you walk away. What a passing mouse cannot do is promise success in people. It moves a target from “no” to “maybe,” never from “maybe” to “yes.” The reasons animal models mislead — different physiology, induced versus spontaneous disease, the dozen ways a mouse is not a small person — get their full treatment when we reach preclinical development (Stage 6, and the modality-specific views like M-AAV-03). For now, plant the flag and remember the empirical backbone behind it.

That backbone is a single, brutal number. Of all the drugs that make it into human trials — every one of which has already cleared its animal models, or it would never have been allowed near a person — roughly 86% still fail (Wong, Siah & Lo, Biostatistics 2019; the likelihood of a Phase 1 drug reaching approval is about 13.8% across diseases, and closer to 3.4% in oncology). Read that the right way. The animal model is not a weak filter that lets a few bad ones through. It is a filter that everything downstream has already passed — and most of what passes it still dies. The mouse was necessary. It was nowhere near sufficient. (For why “86% fail” and “13.8% succeed” are the same fact wearing different clothes, and how to keep numbers like these from fooling you, see F-13.)

The Pyramid You’re Actually Climbing

If a clean mouse isn’t proof, what is? The honest answer is that nothing short of a successful human trial is proof — that is what trials are for. But validation is the work you do before the trial to decide whether the trial is worth running, and not all pre-trial evidence is equal. It helps to picture a pyramid, with the cheap, weak, abundant evidence at the bottom and the rare, strong, expensive evidence at the top. (Treat the pyramid as a teaching frame, not a law of nature — the levels blur, and good programs gather evidence at every tier.)

At the base sits cell-culture work. You take human cells in a dish, knock the target’s gene down with a tool that suppresses it, or knock it out entirely, and watch what changes. This is fast and cheap and you can do it a hundred ways. It is also the furthest from a living person — a cell in a dish is missing the body it’s supposed to be part of, the blood and the neighbors and the time. Cell work tells you the target can do something. It rarely tells you that doing it matters.

A tier up sit animal disease models — the knockout mouse from the opening, and its many cousins. These are more bodylike: a real organism, real organs, real disease over real time. This is genuinely more informative than the dish, which is why every program leans on it. But we have already named its ceiling. It filters; it does not predict. The seduction lives on this tier precisely because the evidence here is good enough to feel like proof while falling short of it.

At the top sits the strongest non-clinical evidence anyone has found: human genetics. And the cleanest version of it is something close to a miracle of free experimentation. Somewhere in the human population, by ordinary chance, are people born with a broken copy of almost any given gene — a mutation that switches the protein off, or turns it down, from birth. Nature has already run the experiment your drug is trying to imitate: it shut the target down, in a human being, for an entire lifetime, with no company and no budget. The question is simply whether anyone thought to go look. These are the “experiments of nature” — naturally occurring loss-of-function variants — and reading them is now possible at scale because of biobanks, the vast linked collections of DNA and health records that let researchers ask, across hundreds of thousands of people, what happens to the ones who were born without this protein?

Two primers sit under this tier. That genes carry instructions a body can run, lose, or mutate — rather than fixed blueprints — is the F-03 idea the whole approach rests on; that trusting an association drawn across that many people takes real statistics is the F-07 one.

Why is the top of the pyramid so much stronger than the rest? Because it answers the exact question validation is asking — what does turning this target off do to a person? — using a person, over a lifetime, instead of a model of one. And the payoff shows up in the only number that matters at the end: drug targets that have human-genetic support are roughly twice as likely to make it all the way to an approved drug (Nelson et al., Nature Genetics 2015). Hold onto that 2x; it is the spine of this stage. It is worth saying honestly what it is and isn’t: later work has refined the figure — King and colleagues revised it in 2019, and a 2024 update in Nature revisited it again — so “2x” is a robust, well-replicated headline, not a constant carved in stone. The exact multiplier moves. The direction does not. Genetic support roughly doubles your odds, and in a business where most things fail, doubling your odds is enormous.

What Nature’s Best Experiment Looks Like

The 2x is an abstraction until you see it do its work, so look at the case that made it famous.

By the early 2000s, a protein called PCSK9 had been nominated as a target for high cholesterol. The pathway made sense; the biology hung together; it was a good nomination. But “good nomination” is exactly where we started this article — a fire truck at a fire. What turned PCSK9 from a plausible bet into one of the most validated targets in modern medicine was not a mouse. It was people.

Researchers went looking, in a large cohort study, for individuals carrying loss-of-function mutations in PCSK9 — people born with a broken copy of the gene, their PCSK9 turned down or off for life. They found them. In one group, about 2.6% of people carried nonsense mutations that crippled the protein, and those people walked around with LDL cholesterol — the “bad” cholesterol that drives heart disease — roughly 28% lower than everyone else, and an 88% lower risk of coronary heart disease (Cohen, Hobbs et al., NEJM 2006). A second, milder variant in another group lowered LDL by about 15% and cut heart-disease risk by roughly 47%. These were observational findings, not a trial — but every figure pointed the same way, and one mattered most for the decision to spend real money: the people missing PCSK9 were healthy. No syndrome, no hidden cost, no compensating disease. Nature had run the knockout in humans, across whole lifetimes, and the verdict was unambiguous — turn this protein down and the disease falls away, and the body is fine without it.

That is what the top of the pyramid looks like. Not a model of the disease, but the answer itself, written in human biology before any drug existed. And the validation paid off the way the 2x predicts it should: the target became drugs that worked. Two antibody drugs that block PCSK9 — evolocumab and alirocumab — were approved in 2015, and a later medicine, inclisiran, that silences the PCSK9 message inside the cell rather than blocking the finished protein, was approved a few years after. Notice that the same validated target supports drugs of completely different kinds — an antibody that grabs the protein, an RNA-silencing therapy that stops it being made (the M-ASO-03 and related modality views walk through how each kind validates a target in its own idiom). Validation is about the target. How you grip it comes later.

It is worth saying that PCSK9 is the dream, not the norm. Most targets do not come with a tidy population of healthy human knockouts waiting in a biobank. Sometimes the cleanest validation is hard-won and arrives crookedly — the cancer-immunotherapy target PD-1 was misunderstood for the better part of a decade before its role came clear (CS-KEYTRUDA-01), and even a beautifully validated driver like the fusion protein behind chronic myeloid leukemia (CS-GLEEVEC-01) had to be proven to be the cause and not merely a passenger. Validation is a spectrum of confidence, not a stamp. The discipline is to know, honestly, where on that spectrum you actually are — and to price the spend accordingly.

How a Program Talks Itself into a Very Expensive Yes

Which brings us back to the room, and the slide, and the clean mouse reproduced across three labs — and to the real hazard of this stage, which is not in the biology. It is in the people.

A team that has spent two years and real reputation getting a target to validation does not arrive neutral. They want the answer to be yes. And confirmation bias — the mind’s habit of weighing the evidence that flatters your hope more heavily than the evidence that threatens it — does not feel like bias from the inside. It feels like being right. The mouse that works gets a celebratory slide. The nagging cell-line result that didn’t replicate gets a footnote, or gets explained away as a bad batch. The literature gets read for support rather than for refutation. None of this requires anyone to be dishonest. It only requires people to be people, with a deadline and a thesis they love.

This is why validation is a discipline and not just a checklist. The structural defense against talking yourself into a yes is to invert the question on purpose: not “what evidence supports this target?” but “what would have to be true for this target to be wrong, and have I gone looking for it as hard as I went looking for the good news?” It is to weight the evidence by where it sits on the pyramid rather than by how exciting it looks on a slide — to treat one population of healthy human knockouts as worth more than a hundred enthusiastic mouse results, even though the mouse results are prettier and there are more of them. The discipline is doing that on purpose. It is, in the end, the same instinct that makes an investor’s diligence memo open with the risks rather than the upside (X-07), and the same gap that the word “translational” papers over when it’s waved around at a conference to mean “we’re sure this will carry over to humans” (X-09). The honest version of translational medicine is this entire pyramid, climbed deliberately, with the 2x in mind and the 86% never forgotten.

So the room should be quieter than the slide wants it to be. A clean, thrice-reproduced mouse is real evidence, and a reason to keep going — to spend on the next tier of validation, to go hunting for human genetic signal, to design the experiments that could still kill the program cheaply. It is not, by itself, a reason to spend forty million dollars as if the human answer were already in. The teams that survive this business are the ones that can hold a beautiful mouse result in one hand and the number 86% in the other, and not let go of either.

When the Knot Is Real

Validation never delivers certainty. What it delivers, done well, is a target you can defend with a causal story rather than a hopeful one — evidence climbed deliberately from the dish to the animal to, when you are lucky, the rare human in whom nature already ran your experiment and the disease fell away. The knot in Stage 2 was a nomination, a bet about cause. The knot after validation is a bet you have tried your hardest to disprove and failed to — which is the strongest thing you can honestly say about a target before a single patient has been dosed.

That is enough to spend on. And the moment it’s enough, the question changes shape entirely. Up to now, everything has been about the target — is it real, does it matter, will turning it off help. None of that work has produced a drug. A validated target is a door you are now certain is worth opening, with nothing yet in hand to open it. The next stage is the search for that something: a molecule — out of a chemical universe almost too large to picture — that actually grips this target and moves it. That is hit discovery, and it is where the search for the drug itself finally begins.

This is the free spine of The Lead Compound.

You’ve just read one stage of how a medicine actually gets made. The spine is the free through-line — the whole pipeline, start to finish. The full course goes deeper: every drug class (antibodies, mRNA, cell and gene therapy, peptides, and more) and the real, documented stories behind the medicines that defined them — Ozempic, Keytruda, Gleevec, Comirnaty, and dozens more.

← Previous (SP-02) · How to read this course · Next: (SP-04) →