Organoids and simulations are entering drug regulation. Validation has not caught up

The phrase "new approach methodologies" is regulatory language, not marketing language, and that is what makes it worth understanding. NAMs is the umbrella term for the human-derived and computational models being positioned to do work that animal testing has done for a century: organoids, microphysiological systems — organ-on-a-chip devices — human stem-cell-derived cell systems, and in silico simulation.

Two review articles published within a fortnight of each other in April, in Science and Cell, set out where that field has reached. A third paper published in July argues that its governance has not kept up.

The problem NAMs are meant to solve

Both reviews open from the same failure statistic rather than from an animal welfare argument.

The Science review states that despite unprecedented technological progress, most drug candidates continue to fail in clinical trials, reflecting a persistent gap between preclinical models and human biology [s1]. The Cell review makes the parallel point that traditional animal-based drug discovery has high failure rates, which has prompted the search for and adoption of human-centred new approach methodologies [s2].

This is the substantive case. A mouse is not a small human, and the compounds that survive mouse pharmacology and then fail in people represent an enormous amount of wasted capital and wasted patient participation. NAMs are proposed as models built from human material in the first place.

What is actually included

The Science review describes NAMs as spanning human-derived cellular systems, microphysiological platforms, and artificial intelligence [s1]. The Cell review is more specific about coverage: stem cell-, organoid-, and in silico-based NAMs, supported by evolving frameworks from the National Institutes of Health and the Food and Drug Administration, now span the entire drug-discovery continuum — from disease modelling and drug design through to efficacy testing [s2].

That range is the field's strongest claim and its biggest evidentiary liability. "NAMs" covers a well-characterised liver-chip assay and a machine-learning toxicity predictor in the same breath, and the two require entirely different kinds of validation.

The regulatory position

The Science review notes that recent regulatory reforms, including the US FDA Modernization Act 3.0, have begun to position NAMs as a complement to or replacement for animal testing [s1]. The Cell review describes NIH and FDA frameworks as evolving in support of these methods [s2].

"Complement to or replacement for" is doing significant work in that sentence. Permitting a sponsor to submit organoid data instead of animal data changes what evidence a regulator receives about a compound before it reaches a human being. That is a consequential shift regardless of one's view of animal research, and it depends entirely on whether the replacing model predicts human response at least as well as the model it replaces.

Both reviews are explicit that this is not settled. The Cell authors examine the key biological, technical and regulatory bottlenecks that need to be addressed to enable robust translational adoption [s2]. The Science authors outline a roadmap for embedding human-based science in a predictive, efficient and ethically grounded infrastructure [s1] — a roadmap being a description of where something is going, not where it is.

The validation gap

The sharpest critique comes from a July paper in Drug Discovery Today, and it concerns the combination that both reviews treat as the field's future: organoids plus AI.

Its argument is that organoid-AI platforms are becoming decision systems in drug discovery rather than merely combined tools — they shape compound prioritisation, toxicity assessment and whether a programme proceeds [s3]. Yet governance often validates the biological model and the computational model separately, even when the evidential claim depends on their interaction [s3].

The failure modes the paper describes are specific. Donor imbalance, batch effects and culture drift in the organoid system can become algorithmic shortcuts — the model learns to predict the batch rather than the biology. In the other direction, confident model outputs can obscure weak biological transportability, meaning a system that performs well on the donors it was built from and poorly on anyone else [s3].

Separate validation, the authors write, can create false assurance [s3]. Each half passes its own test; the combination is never tested as the thing it is being used as.

What is proposed instead

The paper argues for proportionate, platform-level governance organised around a single specified context of use, with linked provenance, transportability testing, and predefined fallback rules, scaled to the stakes of the decision being made [s3].

Unpacked: state exactly what question the platform is being used to answer; track where every piece of biological material and every model input came from; test explicitly whether performance holds outside the source population; and decide in advance what happens when it does not. The authors are clear that the aim is not to slow adoption but to make these platforms credible enough to act as preclinical gatekeepers [s3].

How to read NAM claims

Three questions separate a substantiated claim from a promotional one.

What is the context of use? A model validated for hepatotoxicity screening tells you nothing about cardiac safety, and NAM validation is context-specific by design.

What was it compared against? Replacing animal testing requires demonstrating better prediction of human outcomes, which means benchmarking against human data — not against the animal model, and not against nothing.

Was the whole system evaluated? If an organoid platform and a model were each validated alone and the decision depends on both, the paper's central objection applies [s3].

What to watch

The practical test is regulatory submissions. As sponsors begin filing NAM-derived evidence in place of animal data under the reforms the Science review describes [s1], the question becomes whether the compounds cleared that way behave in humans as predicted. That answer arrives on the timescale of clinical trials, not press releases, and it is the only one that settles the argument.