A NAMs evidence package that shows FDA what the human-based data covers — and the animals it could spare
The case for fewer animal studies, ready to sign in days.
See one case, screen by screen ↓It is the case a sponsor puts to FDA for shortening, replacing or dropping an animal study because new approach methods — human cell assays, organ chips, organoids and computer models — answer what the study was there to answer. NAMs Evidence Package builds it: the coverage matrix, the weight-of-evidence justification, the model credibility assessments and the FDA meeting questions, every sentence cited, for the toxicologist and the regulatory lead to sign.
How to choose AI solutions for life sciences →
Updated
FDA's April 2025 roadmap lets a sponsor shorten or drop an animal study when human-based evidence covers what that study was there to answer — starting with monoclonal antibodies, where the routine 6-month primate study can come down to 3 months. The evidence usually exists: GLP study reports and SEND data, whole-blood cytokine release and tissue cross-reactivity, liver chips and organoids, PBPK and epitope models, human genetics, Phase 1 data, the literature and ICH S6(R1). But it sits in different archives and labs, and the toxicologist and the regulatory writer assemble the weight-of-evidence case by hand.
Nobody can see, on one page, which safety endpoints are covered and which animal cohort each one still holds. So the cheapest gap stays open — a liver chip run on one hepatocyte donor when the qualification plan asks for three — while 48 monkeys stay in the 6-month cohort. And each model that stands in for an animal endpoint has to be assessed for credibility, using the seven steps in FDA's 2025 draft guidance, before a briefing book can lean on it.
Estimated split for one evidence package: about 24 working days by hand (the low end of the 3–6 weeks it takes) and the 5-day target with the solution.
Seven agents read, grade, count, draft and check; a toxicologist and a regulatory lead sign.
Northwind Bio's veltrimab (NWB-2207) is a humanized IgG1 against a cytokine receptor, in Phase 2 for plaque psoriasis. Its remaining animal plan holds 144 monkeys. Here is what happens when the missing liver chip data arrives, screen by screen, in the working solution.
Lena's coverage matrix puts veltrimab's safety endpoints down the side — hepatotoxicity, cytokine release, immune function, immunogenicity, off-target binding, exposure margin, embryo-fetal development — and human cell assays, organ-on-chip, organoids, in silico and supporting human data across. Each cell is graded from its cited evidence. A 144-dot counter shows 32 monkeys could be spared; 48 are still held by hepatotoxicity and 64 by embryo-fetal development.
“Covered = one new approach method Strong and a second line at least Moderate.”
Hepatotoxicity asks whether months 4–6 of dosing could injure the liver in a way a 3-month study would miss. The liver chip result is clean, but LC-0118 used one hepatocyte donor, so the cell is graded Weak and the endpoint is only partly covered. That one cell carries the case for the 6-month chronic cohort: 48 NHPs, none of them spared yet.
LC-0118: “Only one hepatocyte donor was tested in this run; the method's qualification plan calls for at least three donors to capture human variability.”
Dana's lab sends LC-0131, the liver chip validation run. Across three hepatocyte donors — 2 female, 1 male, aged 34–61 — veltrimab at 100 and 1,000 µg/mL caused no change in ALT, albumin, urea or LDH over 14 days. The reference hepatotoxin raised ALT 3.6- to 4.4-fold in all three donors by Day 7, so each chip could detect injury.
The run waits under “Evidence to map”, already matched to Hepatotoxicity × Organ-on-chip.
The NAM assay reader reads LC-0131: 3 donors, 42 endpoints, the reference toxin caught in every chip. The evidence mapper regrades Hepatotoxicity × Organ-on-chip from Weak to Strong, and with the Moderate human hepatocyte and Phase 1 liver test lines beside it, the endpoint is covered. The critic checks the 2 new citations against their source. The counter rises from 32 to 80.
“Hepatotoxicity covered — 48 more monkeys could be spared (80 of 144)”
The package weighs the case on a balance: a clean 4-week GLP study with the NOAEL at the top dose, exposure margins of 38× (AUC) and 52× (Cmax), no cytokine release in 12 human donors, and no liver injury signal in 3 liver chip donors. Weight of evidence reads 88 of 100, a strong case: 8 of 8 FDA expectations met, no gaps, 48 citations checked. One concern is still open — ADA in 2 of 10 high-dose animals — with “Accept the mitigation” beside it.
The in silico immunogenicity model v3.2 goes through FDA's risk-based credibility framework: question of interest, context of use, model risk, credibility plan, execution, documentation, adequacy. Model influence is medium and decision consequence is medium, so model risk is medium. Validated on 217 antibodies with known clinical ADA: Spearman 0.71 and 84% correct classification.
Adequacy: “Adequate for a supporting role. Not adequate as the only immunogenicity evidence.”
The package writer drafts the meeting questions with a sponsor position and citations under each — from whether the 4-week study and the human in vitro data show no concerning signals, to whether the liver chip and whole-blood data may be submitted as supportive human data. For a Type C meeting the request goes in on Oct 20, 2026 and FDA meets within 75 days, by Jan 3, 2027. Omar signed the questions on Oct 6.
Package writer: no sentence without a citation, and no “proves” or “guarantees”.
Before she signs, the checks are on screen: every expectation has evidence, every sentence cited with 48 citations checked by the critic, no overstated wording, model credibility assessed for 2 models, and Omar's regulatory sign-off from Oct 6 at 17:40. Lena signs with her password and the meaning “Approved for submission to FDA”. Priya Raman, the 3Rs officer, is told.
“Package signed by both reviewers — Priya Raman notified”
Teriquamab (NWB-4402) binds a human-specific receptor, so its plan holds a transgenic-mouse study of 120 mice. Kidney injury is covered by a three-donor glomerulus chip and kidney organoids — 84 mice could be spared. Human exposure at first dose still holds 36: the PBPK model's first verification run put 1 of 6 reference antibodies outside its acceptance range, and Lena's review asks Sam Patel for a second run.
TPK-011: “A second run is scheduled.”
The dashboard shows packages signed, the animals that could be spared by programme — agreed with FDA, proposed or still assessing — the median days to a draft ready to sign against about 24 by hand, and the share of claims verified by the critic. A heatmap shows how each package covers FDA's eight expectations. Counts stay “could be spared” until FDA agrees.
The same package, seen by the five people who build, sign and report it — what their weeks looked like, and what they look like now.
Seven agents read, grade, count, draft and check; a toxicologist and a regulatory lead sign.
Reads GLP study reports, SEND datasets and toxicokinetic tables, and extracts design, findings by organ, NOAEL, exposure and ADA with the page and passage for each value.
Reads cytokine-release, tissue cross-reactivity, membrane-protein screens, organ-chip and organoid reports, and checks donors, controls and acceptance criteria.
Finds target biology, class experience, NAM predictivity studies and FDA or ICH statements, and watches for new ones.
Grades each endpoint against each method and each FDA expectation against each evidence line, finds the gaps, weighs the case and counts the animals each covered endpoint could spare.
Runs FDA's seven-step, risk-based credibility framework for each computational model whose output stands in for an animal endpoint.
Drafts the justification and the FDA meeting questions, with a sponsor position under each question and a citation on every sentence.
Checks every number and quote against its source and every claim against the evidence matrix before a person sees the draft.
Both sign the package. The agents propose; a named person decides.
Ask in plain words — or tell it what to change. Every answer cites the passage behind it.
What still holds Veltrimab monkeys in the plan?
Of 144 monkeys in the remaining Veltrimab plan, 32 could be spared today. Hepatotoxicity holds 48 in the 6-month cohort: the liver chip used one donor and three are needed. Embryo-fetal development holds 64 in the ePPND study: the placenta chip has not been run. The three-donor liver chip run is waiting in Evidence to map — adding it would cover hepatotoxicity.
Why is Veltrimab ready to ask for a 3-month study?
The 4-week GLP study found no adverse effects up to 100 mg/kg, with margins of 38× (AUC) and 52× (Cmax). Human-based data agree: no cytokine release in 12 donors, and only expected binding in 36 human tissues. That fits FDA's roadmap for antibodies.
What is still weak in the Teriquamab package?
PKG-0409 has 3 gaps out of 8 expectations: organ toxicity (liver chip run not yet done), immunogenicity (DC–T-cell assay) and target biology (human genetics review). Because the target is human-specific, FDA's roadmap names this case — a human in vitro battery plus a PBPK model instead of a transgenic mouse. The PBPK model is high risk under the credibility framework because it carries the conclusion alone.
Add a rule: every cytokine-release assay needs a positive control
Done. A cytokine-release assay without a responsive positive control is now graded as a gap. All 3 cytokine assays in the open packages have one — anti-CD3 responded in 12 of 12 donors, for example — so no grades change. The rule is now in Settings.
13 screens from the working solution, on its sample data. Pick one to see it large.
Who signs and the rules the agents grade by are settings. Turning the two-lines or three-donor rule on or off re-grades the packages it affects, and the change goes on the audit trail.
| Setting | Default | Choose from |
|---|---|---|
| Scientific sign-off | Dr. Lena Ortiz | Dr. Lena Ortiz · Dr. Maya Chen |
| Regulatory sign-off | Omar Haddad | Omar Haddad · Sam Patel |
| Tell when signed | Priya Raman · 3Rs officer | Priya Raman, 3Rs officer · Nobody |
| Meaning of signature | Approved for submission to FDA | Approved for submission to FDA · Reviewed, not yet for submission |
| Two independent lines of evidence per expectation | On | An expectation is Met only when two lines agree |
| Flag human in vitro results from fewer than 3 donors | On | Single-donor organ-chip or cytokine data are graded as a gap |
| Credibility assessment for every model standing in for an animal endpoint | On | FDA's 7-step risk-based framework, model risk shown |
| Block overstated wording | On | The critic removes words like “proves” or “guarantees” |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample programme data · “target” = the design goal, measured in the live solution · “estimated” = our estimate · “typical” = published figures (FDA, Roadmap to Reducing Animal Testing in Preclinical Safety Studies, April 2025 — about 144 non-human primates in a typical antibody programme, up to $50,000 each, and the 6-month primate study reduced to 3 months; FDA draft guidance on AI and model credibility, 2025). People, companies, programmes and studies named on this page — Northwind Bio, veltrimab, teriquamab — are fictional: characters and sample data in the working solution.
It is the case a sponsor puts to FDA for shortening, replacing or dropping an animal study because new approach methods — human cell assays, organ chips, organoids and computer models — answer what the study was there to answer. NAMs Evidence Package builds it: the coverage matrix, the weight-of-evidence justification, the model credibility assessments and the FDA meeting questions, every sentence cited, for the toxicologist and the regulatory lead to sign.
The roadmap is one of the sources the agents cite. For a monoclonal antibody with no concerning signals in its 1-month study plus NAM tests, the package can ask to shorten the 6-month primate study to 3 months; for a human-specific target, it can propose a human in vitro battery plus a PBPK model instead of a transgenic-mouse study. The regulatory watch maps new FDA and ICH statements to open packages.
Each safety endpoint is tied to the cohort in the remaining plan that exists to answer it — the 6-month chronic cohort, a cytokine satellite, the ePPND study. When an endpoint is covered, its animals count as could be spared; when evidence is partial or missing, they stay held. The counts stay “could be spared” until FDA agrees.
One new approach method graded Strong and a second, independent line at least Moderate. A human in vitro run from fewer than 3 donors, or with a failed positive control, is flagged and cannot carry the case — which is why a single-donor liver chip run holds the 48-monkey cohort until the three-donor run arrives.
Yes. Every model that stands in for an animal endpoint goes through FDA's seven-step, risk-based credibility framework: question of interest, context of use, model risk, credibility plan, execution, documentation and adequacy. Model risk comes from model influence and decision consequence, and is never lowered without a recorded reason.
The package writer cites every sentence and blocks wording such as “proves” or “guarantees”. The critic checks every number and quote against its source before a person sees the draft, and sends the draft back if any citation fails.
Yes. The agents read, grade, count and draft; they never sign and never file. The programme toxicologist signs the science and the nonclinical regulatory lead signs the meeting questions, each with the meaning recorded, and both signatures are needed before a package goes to FDA.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your study reports, assay and model reports, and the evidence rules you grade by — and it goes live once every quality gate has passed. We will walk you through it on one of your own programmes first.
Sourced answers from papers, patents and reports in seconds, our own disagreements first.
1-min demoCompetitive Pipeline TrackingEvery rival trial, readout and approval in a sourced brief each Monday
1-min demoPreclinical Study Report AutomationA draft GLP report in hours, every number traced to raw data.
1-min demoWe'll run NAMs Evidence Package on the study reports, assays and models of one of your own programmes.