Experimental Therapy Testing: How to know if a therapy is safe and effective before it reaches a patient

For many families, a genetic diagnosis is supposed to be the finish line. Years of appointments. Specialist after specialist. A test that finally puts a name to it.

Then the name arrives, and a harder question takes its place.

If there might be a treatment, how soon can we try it?

It is the most human instinct in the world. A child is sick, a fix might exist, so why wait? The answer is that an untested therapy can fail in two directions. It can do nothing. Or it can cause harm. Before any new therapy goes into a person, the work is to find out which, while the only thing at risk is a model and not a patient. That is what makes an experimental therapy responsible to attempt.

The two questions every experimental therapy has to answer

Strip away the jargon and preclinical testing comes down to two plain questions.

You answer both the same way. You test the therapy in models that stand in for the patient, long before the patient is ever involved. The whole discipline of preclinical research exists to answer these two questions honestly, so a regulator like the FDA, and a family, can decide whether a first human dose is justified.

How do you choose the right model?

A model is a living system, or a set of cells, that stands in for the patient’s biology so an idea can be tested safely. The choice matters enormously, because not every model can answer both questions, and not every organism models every gene.

Two things make a model good enough to use.

First, it has to carry the patient’s actual mechanism. A loss-of-function variant, a toxic gain-of-function variant, and a splicing error are three different problems, and a model that mimics one tells you nothing reliable about the others.

Second, it has to give a signal you can measure. Picture a dimmer switch. If the light is barely on to begin with, you cannot tell which switch actually dims it. A screen works the same way. You need enough of the gene’s activity present to see whether a candidate moved it.Knowing which models even exist used to be the hard part. MARRVEL, developed by researchers at Baylor College of Medicine, was built to fix that. It connects human genetic findings with model organism data, showing in one place where a gene has already been studied across species, which tissues it affects, how conserved it is, and whether there is an ortholog you could test in. [1] That is how a team decides, before spending a dollar at the bench, which models are worth pursuing.

Patient cells show whether a therapy works on the exact variant. Whole animals show whether it is safe and effective across a living system.

The two often work together. A patient's own cells, grown in a dish and carrying the precise change, are usually the cleanest test of whether a candidate corrects the underlying problem. Animal models then add what a dish cannot: whether the same effect holds in a living body, whether the delivery is tolerated, and whether anything harmful shows up at the doses a person would actually receive.

Where this shows up in the therapies that worked

This is not theory. Every custom therapy in the n-of-1 roadmap we have written about cleared this bar before reaching a patient.

Milasen. When Tim Yu's team designed a splice-correcting therapy for a single child with a rare form of Batten disease, they did not start in the patient. They tested the candidate in her own cells to confirm it corrected the defect, and ran safety checks, before she was ever dosed. [2]

Baby KJ. The personalized gene-editing therapy made for an infant with a severe metabolic disorder was validated in laboratory cells and animal models carrying the relevant change. The team could see the edit work, and watch for safety signals, before it reached the baby. [3]

The pattern repeats every time. Before any of these therapies touched a person, the same two questions had already been answered in a carefully chosen model. That work was not a delay before the real thing. It was the thing that made the real thing possible.

Where this is heading: from animals to computer models

For most of modern drug development, answering the safety question meant animal studies. That is starting to change, and the change matters for rare disease most of all.

In March 2026, the FDA's drug center released draft guidance encouraging developers to use what it calls new approach methodologies: human cell systems, organs-on-chips, and in silico computer models, including AI, to predict whether a therapy is safe, in some cases in place of animal testing. [4] The agency did not just permit these tools. It noted that several validated versions have already outperformed traditional animal models at predicting how a drug behaves in people, and it qualified its first AI-based drug development tool the same year. [5]

A computer model can fold in everything already known about a gene and a variant, and predict an effect in a fraction of the time an animal study takes.

For a family facing a rare disease, the appeal is concrete. These methods are faster, cost less, and are built directly on human biology rather than a stand-in species. A model that takes months and a colony of animals could, increasingly, be a simulation that runs on existing human data. For ultra-rare conditions, where there may be no good animal model and no time to build one, that shift could be the difference between a therapy that is feasible and one that never gets attempted.

This is still early. The guidance is in draft, the methods still need validation, and cell and animal models are not disappearing tomorrow. But the direction is set. The way we prove a therapy safe and effective is moving toward evidence that is human-relevant, computational, and fast, which is exactly the kind of infrastructure personalized medicine needs to reach more patients.

So, what now?

Knowing whether an experimental therapy is safe and effective is not a matter of conviction. It is a matter of evidence, gathered in the right models, in the right order. Choose models that answer both questions and a genetic finding becomes a therapy a family can responsibly pursue. Skip that work and even a brilliant idea has no business going near a person.

At Nome, deciding what to test in is one of the first questions we work through after a diagnosis. We look at what is already known across model organisms, weigh a patient's own cells against animal models, and increasingly bring computational tools to bear on the same questions. The goal is to choose the systems that can show both whether a therapy works and whether it is safe. That evidence is what lets a custom therapy advance under the FDA's Plausible Mechanism Framework, and what lets a family move forward with their eyes open.

About Nome

This article is for informational purposes only and is not medical advice. Families should consult their care team about decisions related to diagnosis and treatment.

References

Wang J, et al. "MARRVEL: Integration of Human and Model Organism Genetic Resources to Facilitate Functional Annotation of the Human Genome." American Journal of Human Genetics, 2017. https://pmc.ncbi.nlm.nih.gov/articles/PMC5670038/

Kim J, et al. "Patient-Customized Oligonucleotide Therapy for a Rare Genetic Disease." New England Journal of Medicine, 2019.

Musunuru K, et al. "Patient-Specific In Vivo Gene Editing to Treat a Rare Genetic Disease." New England Journal of Medicine, 2025.

"General Considerations for the Use of New Approach Methodologies in Drug Development." Draft Guidance for Industry, FDA Center for Drug Evaluation and Research (CDER), March 18, 2026. https://www.fda.gov/news-events/press-announcements/fda-releases-draft-guidance-alternatives-animal-testing-drug-development

"FDA Achieves Year 1 Goals in Reducing Animal Testing in Drug Development." FDA, April 2026. https://www.fda.gov/news-events/press-announcements/fda-achieves-year-1-goals-reducing-animal-testing-drug-development