A target safety assessment that shows, organ by organ, where the knockout mouse and people disagree
A cited safety assessment for any new target in under two days, every risk scored.
See one case, screen by screen ↓Before chemistry starts on a new target, a discovery toxicologist has to say what switching it off could do to the body. That means reading knockout and conditional mouse phenotypes, loss-of-function constraint and biobank carrier scans, tissue expression in people and in the toxicology species, the labels and adverse-event reports of drugs on the same pathway, and the company's own early studies — then turning it into a ranked list of liabilities, each scored for probability and impact, with the studies that would settle it.
Each evidence line is read on its own, so the places where the mouse and people disagree surface late, often in the GLP study. And the mouse misleads in both directions: Hmgcr and Ptgs2 knockouts are lethal, yet statins and coxibs are approved drugs, and a QT signal in a mouse — whose QT barely changes with heart rate — needs a non-rodent check.
Estimated split for one target assessment of typical scope, by hand and with the solution — about two weeks by hand.
Nine agents plan, read six evidence lines in parallel, light the body map organ by organ, score, draft and challenge; a toxicologist re-scores with a reason and signs.
Dr. Lena Ortiz, a discovery toxicologist at Northwind Bio, is preparing SRK9 for the portfolio committee on Oct 22. This morning Dr. Maya Chen asked for one more: NLRP15, for gout. Here is her day, screen by screen, in the working solution.
Her home screen is the work itself: SRK9, signal-regulating kinase 9, for rheumatoid arthritis, oral small molecule NWB-5521 — 10 liabilities from 16 key sources. Each of fourteen organ systems is coloured by its highest risk; two are high, serious infections and embryo-fetal lethality. On blood and bone marrow, heterozygous knockouts have 22% fewer neutrophils, but carriers have normal counts.
“The mouse shows it; people with a broken copy do not. Weigh the human evidence and settle it in a second species before first-in-human.”
Every organ carries two marks — M for the knockout mouse, H for people with a broken copy: filled shows the finding, open is clear, dashed is no data. Coloured by agreement, the SRK9 body splits four ways: mouse and human disagree on the liver and blood, both show infection and the embryo, four organs have no human data and six are clear in both.
The contradictions are where the toxicologist's judgement is needed — so they are the first thing she sees.
The assessment planner resolves the symbol and starts six evidence agents in parallel. Findings land with their organ and source: people with one broken copy have 31% lower gout risk, which supports the target; knockout mice are viable and fertile; NLRP15 is high in colonic epithelium and the knockouts get worse chemically induced colitis.
“Six evidence agents are reading NLRP15 · 728 of 1,214 papers · 4 of 14 organs scored”
Knockout telemetry shows a QT interval 14 ms longer than littermates (p = 0.01, 12 mice per group), with no arrhythmias. But 1,206 people with a broken copy and a resting ECG show QTc +0.6 ms (95% CI −1.1 to 2.3) and no excess of arrhythmia or syncope. Seven liabilities: none high, three medium, four low; mouse and human disagree on three organs.
“NLRP15 assessed in 9 minutes of agent time · 7 liabilities · 0 high · 3 medium · 4 low”
The liability “QT prolongation seen in mice only” is scored Unlikely × Serious, medium, at 71% agent confidence. Its evidence ladder shows one of six lines supporting the risk and one arguing against, each with the cited passage beside it — the biobank ECG scan and the mouse telemetry paper. The critic challenged three NLRP15 liabilities, the heart among them.
Critic: “Mouse QT is a weak guide to people: it barely changes with heart rate. The critic suggests keeping probability at Unlikely, weighting the carrier ECGs, and asking for a non-rodent readout.”
The drafter writes the assessment section by section — summary and recommendation, target and pathway, human genetics, animal models, expression, same-mechanism drugs, liability register, monitoring and de-risking plan — with every statement tied to a numbered source. Lena can export it to Word or send it to the IND summary.
The report drafter's rule: no statement without a citation.
The agents suggested three studies, each tied to the liability it answers: an ex vivo monocyte infection-response assay (3 weeks, $28k), colon histology and calprotectin in the 4-week rat study ($7k), and dog telemetry in the 4-week study with a hERG patch clamp (4 weeks, $34k, Priya Raman). Lena switches the dog study on: three of three in the plan, $69k, longest four weeks.
“Settles the mouse-only QT finding in a species whose ECG behaves like ours.”
The sign dialog shows 7 liabilities, 0 high and 3 studies going out with the report. Of the four recommendations — “Go — no plan needed”, “Go with a de-risking plan”, “Hold for more evidence”, “Stop” — she chooses the second and signs with her password. With no high risk, no second signature is needed.
“By signing I confirm I have reviewed the evidence and approve this target safety assessment. Meaning of signature: approved by toxicologist.”
The review record shows who did what, people and agents alike: the planner started six evidence agents, the risk scorer scored 7 liabilities, the critic challenged 3 and the agent agreed on 2, the drafter drafted the report, Lena added the dog telemetry study and signed. For SRK9, with two high risks left, the same signature routes to Dr. Omar Haddad, head of discovery safety, to co-sign.
“Every agent step and every decision on NLRP15 — immutable, people and agents alike.”
The same assessment, seen by the five people who carry it — what their weeks looked like, and what they look like now.
Nine agents plan, read six evidence lines in parallel, light the body map organ by organ, score, draft and challenge; a toxicologist re-scores with a reason and signs.
Turns a request into an evidence plan: resolves the target symbol, its aliases and paralogs, sets the indication and modality, and starts the six evidence lines in parallel.
Reads loss-of-function counts and constraint, biobank carrier scans, Mendelian disease and case reports — what happens to people born with less of the target.
Collects knockout, knock-in and conditional phenotypes, including embryonic lethality, from the mouse phenotype database and the literature.
Summarises where the target is, by organ system, in human tissues and cells and in mouse, rat, dog and monkey, and flags species where the pattern differs.
Finds drugs on the same target or pathway and extracts every safety finding from labels, boxed warnings, public adverse event reports and trial results.
Checks paralogs, selectivity panels, hERG and secondary pharmacology, and what the company's own toxicology archive already showed.
Merges findings into liabilities by organ system, scores probability × impact and monitorability, sets the mouse-versus-human verdict and suggests de-risking studies.
Writes each section of the house template from the liability register and the sources, keeping citations as numbered references.
Challenges single-source liabilities, species relevance such as a mouse QT finding, and developmental findings filed under the wrong organ.
Re-scores with a reason and signs. The agents propose; a named person decides.
Ask in plain words — why a risk is high, where mouse and human disagree, what changed since signing — or tell it what to change. Every answer points to its source passage.
Why is SRK9 a high infection risk?
Because four evidence lines agree: people with no SRK9 have recurrent pyogenic infection in 3 families; myeloid knockout mice carry a 40-fold bacterial burden; SRK9 is highest in neutrophils and monocytes; and a same-mechanism drug has serious infections at 3.1 per 100 patient-years. What argues for a window: 418 people with one broken copy are healthy.
Where do mouse and human disagree on SRK9?
On 2 organs: blood and bone marrow, and the liver. Heterozygous knockouts have 22% fewer neutrophils and fatty liver on a high-fat diet, but 418 carriers have normal counts and no liver disease. They agree on the two high risks — infection and the embryo.
Is the NLRP15 heart finding real?
Probably not in people, but it needs one study. Knockout mice have a QT interval 14 ms longer than littermates, while 1,206 carriers with a resting ECG show QTc +0.6 ms and no arrhythmia excess. Mouse QT barely changes with heart rate, so it is a weak guide. Scored Unlikely × Serious (medium); the plan offers dog telemetry and a hERG patch clamp — 4 weeks, $34k.
What changed for ACKR7 since it was signed?
ACKR7 was signed on Aug 30. On Oct 6 the evidence watch found a label update for a same-pathway antibody: drug-induced lupus in 0.3% of patients, reversible after stopping. Knockout mice develop anti-nuclear antibodies at 9 months, so an autoimmunity liability (Unlikely × Dose-limiting) was added and the assessment re-opened. Nothing else changed.
13 screens from the working solution, on its sample data. Pick one to see it large.
The template, the scoring threshold, the sources and who signs are settings — changed in one place and recorded.
| Setting | Default | Choose from |
|---|---|---|
| High risk means | Probability × impact of 12 or more | 10 or more · 12 or more · 15 or more |
| A high risk needs a second signature | On · Dr. Omar Haddad, head of discovery safety | On or off · second signer chosen from the team |
| Accept low risks in one step | On | On · Off — low risks reviewed one by one |
| Critic challenges every single-source liability | On | On · Off |
| Re-scores and dismissals need a reason | Required | Locked on |
| Optional report sections | Species relevance and study design on; competitor safety comparison off | Each on or off; sections 1–8 always included |
| Preprints | Off | On — labelled “not peer reviewed” wherever cited · Off |
| Re-check signed assessments for new evidence | Weekly | Weekly · Every two weeks · Monthly |
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 data · “target” = the design goal, measured in the live solution · “estimated” = our estimate · “typical” = published figures (Cook et al., Nat Rev Drug Discov 2014, AstraZeneca pipeline review 2005–2010: safety, mostly organ toxicity, behind more than half of project failures and 62% of Phase 1 failures; Minikel et al., Nature 2020, on human loss-of-function as a natural knockout and the Hmgcr and Ptgs2 examples). People, companies, targets and molecules named on this page are fictional — characters and sample data in the working solution.
It is the discovery-toxicology review of what switching a target off could do to the body, written before chemistry commits to it: a ranked list of on-target liabilities, each scored for probability and impact, with follow-up studies. Target Safety Assessment builds it from knockout phenotypes, human loss-of-function genetics, expression, same-mechanism drugs, pathway and your own studies, for a toxicologist to review and sign.
Every organ carries two marks — what the knockout mouse shows and what people with a broken copy show — and a verdict: agree, disagree, no human data or no mouse data. The organ panel puts the two findings side by side, cited, with what to do about a disagreement, and one switch recolours the whole body by agreement.
From the sources you allow: licensed literature, the mouse phenotype database, a public exome database, population biobank scans, human and four-species expression atlases, drug labels, public adverse event reports, trial registries and your own toxicology archive. Preprints are off by default and labelled “not peer reviewed” when switched on.
The risk scorer proposes probability × impact on a 5 × 5 matrix — Remote to Expected, Minimal to Programme-stopping — with monitorability and suggested studies. You set what counts as high. The toxicologist accepts, re-scores with a reason or dismisses each one.
No. The agents propose scores and the critic's challenges are shown, never silently applied. A discovery toxicologist decides every liability and signs the assessment with its recommendation, and the head of discovery safety co-signs when a high risk remains.
Signed assessments are re-checked for new labels, adverse-event reports and knockout papers — weekly by default. When new evidence would change a score, the assessment is re-opened and its owner gets an item to review.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your report template, scoring rules and a few past assessments — and it goes live once every quality gate has passed. We will walk you through it on one of your own targets 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 Target Safety Assessment on one of your own targets, against an assessment you have already written.