"Preclinical" means a stage of studies, not a computational result
The word causes more confusion than any other in this field, so it is worth fixing first. FDA's own description of the drug-development process puts preclinical research at step two and defines it by its methods: before testing a drug in people, researchers must find out whether it can cause serious harm, and the two types of preclinical research are in vitro and in vivo. The safety-critical studies among them fall under Good Laboratory Practice.
US regulation draws the boundary tighter still. A "nonclinical laboratory study" is an experiment in which test articles are studied prospectively to determine their safety — and the definition explicitly excludes "basic exploratory studies carried out to determine whether a test article has any potential utility". Screening and discovery work, in other words, is not the regulated nonclinical package; the GLP-governed safety studies are.
Industry speech uses the word differently again: "reaching preclinical" usually means nominating a candidate, the point at which one molecule is chosen to carry the programme. Two readers can therefore hear two different milestones in the same sentence. Any claim about the time it takes to "reach preclinical" should say which one it means.
Target selection and ligand discovery
A programme's ceiling is set at target selection. The target has to be present on the tumour, absent or tolerable on the healthy tissues the construct will reach, and expressed densely enough that a molecularly delivered dose is meaningful. Because expression in healthy tissue travels with the target, no amount of ligand optimisation fixes a bad choice here.
Ligand discovery then proceeds by one of two routes, or both. Selection-based methods — phage display and mRNA display — enrich binders out of very large libraries by experiment. Computational methods generate candidates for a chosen binding site, including de novo design of peptides, mini-proteins and nanobodies, and rank them before anything is synthesised. The modality adds a constraint the general protein-design field does not have: the construct also needs a place to attach a chelator without destroying its binding.
Whichever route produced the candidates, the same in-vitro checkpoint follows: direct binding measurement against the target, usually by SPR or BLI, plus binding on cells that express the target and counter-screens against close homologues. This is where a predicted affinity becomes a measured one — and where the difference between Kd, Ki and IC50 starts to matter for how the result is reported.
In-vivo imaging and biodistribution studies
The first whole-body answer comes from animals, and for this modality it arrives unusually early, because the construct can be imaged. A labelled candidate is administered to tumour-bearing animals and its distribution followed over time, giving tumour uptake, healthy-organ uptake and the clearance route in one experiment.
FDA's guidance for oncology therapeutic radiopharmaceuticals is specific about what this work needs to produce. A single-dose animal biodistribution study informs dose selection for the human biodistribution and dosimetry study; the entire radioactive decay cascade should be considered; organ uptake is extrapolated to human absorbed dose; and the dosimetry methodology and software — including version — are described in the application. The same guidance expects proof-of-concept studies showing tumour uptake and antitumour activity before a first-in-human study.
Two of its conclusions are counter-intuitive and worth knowing. A general toxicology study with the radioactive product itself is usually not warranted — the biodistribution study with added safety endpoints, read against the radiation-toxicity literature, generally suffices. But a general toxicology study of the non-radioactive ligand in one relevant species is expected before the first human study. And genotoxicity, reproductive-toxicity and carcinogenicity studies are not warranted, because ionising radiation is inherently genotoxic and that risk is handled in labelling instead.
The IND-enabling package and its regulatory touchpoints
An investigational new drug application is what permits a human study in the United States. Its content requirements are set out in regulation: adequate pharmacology and toxicology information on the basis of which the sponsor has concluded it is reasonably safe to proceed, with a GLP compliance statement for each nonclinical study, and a chemistry, manufacturing and controls section whose depth scales with the phase.
Radioactive drugs carry one extra clause, and it is the one that shapes the whole programme: the application must contain sufficient data from animal or human studies to allow a reasonable calculation of radiation-absorbed dose to the whole body and critical organs, and Phase 1 studies of radioactive drugs must themselves generate data adequate for dosimetry. FDA's 2025 draft guidance on dosage optimisation asks that dosimetry studies be performed early in clinical development for every novel molecular entity.
The touchpoints, in order:
- Pre-IND meeting — not required, but the point at which the reviewing division, the species, the dosimetry method and the manufacturing expectations get agreed.
- IND submission — the 30-day clock in 21 CFR 312.40: the application takes effect thirty days after FDA receives it unless the agency imposes a clinical hold, and no investigational drug may be given to a human subject before it does.
- Institutional review board approval and informed consent, which are separate from the IND and run in parallel.
- A radioactive-materials licence or licence amendment from the Nuclear Regulatory Commission or an Agreement State, without which the site cannot administer the material.
- In the European Union, a clinical trial application under Regulation 536/2014 with its own assessment clock, an investigational medicinal product dossier whose quality guideline explicitly covers radiolabelled substances, and GMP Annex 3 for manufacture.
Where in-silico design and de-risking sit
Computational work sits before the first wet experiment and alongside every later one. Before: target triage, candidate generation, affinity ranking, off-target screening and biodistribution prediction, none of which consume material. Alongside: each subsequent measurement is a test of a prediction, and the disagreements are where a model earns or loses trust.
What it does not do is move a regulatory gate. ICH S9, the anticancer nonclinical guideline, states that radiopharmaceuticals are not covered by it although some of its principles could be adapted; EMA's dedicated non-clinical guideline for radiopharmaceuticals has been at draft stage since 2018. In a field where the written expectations are this thin, sponsors do not get credit for in-silico evidence — they get a better-chosen molecule to take into the studies the regulator does expect.
Epitorix states its own scope as reaching "4–6 weeks from target to a de-risked computational lead", and the page that publishes it defines the endpoint in the same block: "Epitorix estimates. 'Preclinical' means a designed, ranked and de-risked candidate ready to enter preclinical testing; in vitro and in vivo studies, GLP toxicology, GMP manufacturing and regulatory review follow before any first-in-human study." That sentence is the whole of the claim, and it is worth reading twice — the four to six weeks buy a better starting molecule, not a shorter regulatory path. The comparison it sits in is on the science page; the workspace is EPiQ™.
