An antibody–drug conjugate knowledge base that checks every new design against every construct built before it
Every antibody–drug conjugate the team has built, searchable in plain words, so each new design is checked against every one before it.
See one case, screen by screen ↓Every antibody–drug conjugate a team builds leaves its lessons in a dozen places: the conjugation and characterisation report, the HIC and SEC tables, the repeat-dose study's ophthalmic section, the xenograft study, the decision record that stopped the programme. Inherited programmes add another team's templates and another team's words — “conjugate load”, “monomer loss”, “ocular surface finding”.
So when a new design comes to the board, the prior-art question — has anyone here tried this chemistry, and what happened? — is answered from memory and from whoever is still around. A hydrophobic linker at a ratio near 8 on a topoisomerase-1 payload can come back as a fresh proposal, two programmes after the same chemistry was stopped for eye findings and for aggregate that would not release.
Estimated split for checking one new design before the design board, by hand and with the solution. Hours.
Six specialist agents read, extract, map and check every report, then answer questions and check new designs; the design board chair decides.
NB-ADC-119 is a second run at the NRX-7 target with a topoisomerase-1 payload, and the design board meets on Thursday. Here is what happens before it, screen by screen, in the working solution.
Maya opens the solution to 14 constructs in the registry, 4 facts waiting for a check and 23 designs checked before the board this year. The design space puts drug-to-antibody ratio across the page and payload class down it, coloured by outcome. Under “Needs your check”: NB-ADC-119 — no prior-art check yet. The design board meets on Thursday.
What the portfolio has taught us: “A hydrophobic payload carried at a ratio near 8 pushes aggregate past the release limit — the PEG-extended linker at the same ratio does not.”
She asks which topoisomerase-1 payload constructs had eye findings, and what they had in common. The agents find the five constructs, read the ophthalmic sections of five repeat-dose studies, compare linker, conjugation, ratio and aggregate, and write the answer from nine facts. Two of the five: NB-ADC-103, grade 2–3 corneal findings in 4 of 6 animals at 10 mg/kg, and NB-ADC-106, grade 2 in 3 of 6 at the same dose — both on the hydrophobic tetrapeptide linker, stochastic cysteine, a ratio near 8.
“The three clean constructs all use the PEG-extended linker at a ratio of 4, conjugated site-specifically or enzymatically.”
Every number in the answer carries a citation. Opening the first one brings up TOX-2609, page 78, §7.3 Ophthalmic examinations, with the passage highlighted: corneal epithelial findings in 4 of 6 animals, partly reversible, two animals still affected at day 56, none at 3 mg/kg. From there, the original record is one more click.
The answer marks one fact as still waiting: “no ocular findings” on NB-ADC-118, extracted from today's two-week study, does not carry the conclusion yet. On the construct record, three facts wait for Dana — aggregate at release 1.3% by SEC, 91% of the material at DAR 4 by HIC, no ocular findings up to 12 mg/kg — each beside the page and section it came from, with Confirm and Correct.
Confirm a fact and it becomes searchable; correct it and the correction is on the record.
“A second run at NRX-7 with the payload that gave the strongest growth inhibition in the portfolio.” IgG1, humanised; topoisomerase-1 payload; cleavable tetrapeptide (hydrophobic) linker; stochastic cysteine conjugation; drug-to-antibody ratio 8.0. The check will read 14 confirmed constructs, 186 reports and the stop decisions behind each programme, under three rules set in Settings.
The prior-art reasoner reads the design, scores it against every construct, pulls six reports from the ophthalmic and characterisation sections of the close ones, matches three chemistry rules and drafts three changes. Then it stops and waits for the board chair. The verdict is plain.
“This design repeats a stop we have already paid for. Two constructs with this exact chemistry were stopped — one on safety, one on manufacturability.”
NB-ADC-103 is 91% close and stopped on safety; NB-ADC-106, 88% close on the same target, stopped on manufacturability. Four risks follow, each with its source: ocular findings likely at this ratio and linker (TOX-2609 · p. 78); aggregate of 6.4% and 5.8% against a 3.0% release limit, which three conjugation campaigns did not fix (CMC-0906 · p. 22); NRX-7 responding only above roughly 100,000 copies per cell; free payload above 0.35%.
Every risk names its precedent and its source. A score is never returned without the evidence.
Three changes, each with what it is worth: use the PEG-extended cleavable linker (−27) — none of the four constructs on it had ocular findings; bring the ratio to 4 (−18) — all seven constructs between 3.8 and 4.2 sit inside the release limit; conjugate site-specifically (−12) — 91% of the material at a single ratio on all five site-specific constructs. Maya can apply one at a time or all three and re-score.
With all three changes, repeat risk falls to 21: “Nothing in the portfolio argues against this design.” It now sits with the topoisomerase-1 constructs that showed no eye findings and released well inside the aggregate limit; the antigen-density question goes with it as a patient-selection question. Maya sends it with the check attached, and Priya accepts it for the 8 October board.
The check proposes; the board chair accepts it or sends it back, and either way it is on the record.
The same design, seen by the five people who carry it — what their week looked like, and what it looks like now.
Six specialist agents read, extract, map and check every report, then answer questions and check new designs; the design board chair decides.
Reads study reports, conjugation records and papers, and works out which construct and which study each one is.
Pulls the construct-level facts out of each report: target, format, linker, payload class, conjugation, ratio and distribution, aggregate, free payload, efficacy and findings.
Maps inherited and external naming onto the house terms, keeping the original wording on the record.
Compares every new fact with what the registry already holds for that construct and raises the disagreements.
Scores a new design against every construct built before it, names the nearest precedents and the risks they carry, and proposes changes the portfolio shows work.
Answers questions in plain words from confirmed registry facts, with a citation on every claim.
Accepts or sends back each check. The agents propose; a named person decides.
Ask about any construct, payload, linker or finding — or tell it what to change. Every answer is cited.
Which topoisomerase-1 constructs had eye findings?
Two of five: NB-ADC-103 (grade 2–3 in 4 of 6 animals) and NB-ADC-106 (grade 2 in 3 of 6), both on the hydrophobic tetrapeptide linker at a drug-to-antibody ratio near 8. NB-ADC-109, NB-ADC-112 and NB-ADC-118 use the PEG-extended linker at a ratio of 4 and had none — NB-ADC-109 up to 12 mg/kg in the four-week study, the other two in two-week pilots.
Has anyone here made an ADC against NRX-7 work?
Twice tried, never landed. NB-ADC-114 reached only 48% tumour growth inhibition in a model carrying 40,000 copies per cell; NB-ADC-106 reached 79%, but only in the antigen-high model, and was stopped at the manufacturability gate. NRX-7 needs roughly 100,000 copies per cell before this chemistry works, so a third attempt is a question about patient selection as much as about the molecule.
Check NB-ADC-119 against everything we have built
Running the prior-art check on NB-ADC-119 — NRX-7, topoisomerase-1 payload, hydrophobic tetrapeptide linker, stochastic cysteine, ratio 8. I will score it against all 14 constructs and bring back the precedents, the risks and what to change.
Flag any proposal with a hydrophobic linker above a ratio of 6
Added. From now on every design proposal with a hydrophobic linker above a drug-to-antibody ratio of 6 is flagged before the board, with NB-ADC-103 and NB-ADC-106 attached as the precedent. The rule is on the Settings page and applies to the next check.
13 screens from the working solution, on its sample data. Pick one to see it large.
How facts are confirmed, who signs and what every check must flag are settings, not a project.
| Setting | Default | Choose from |
|---|---|---|
| Answer only from confirmed facts | On | On · Off |
| Confidence floor — below it, the answer says the evidence is thin | 0.85 | 0.75 · 0.85 · 0.90 |
| Include external literature, always labelled | On | On · Off |
| Who confirms what | Knowledge steward, toxicologist, CMC scientist | Assigned by name |
| Attach a check to every proposal | On | On · Off |
| Flag a payload class with a known organ finding | On | On · Off |
| Flag a target with a failed earlier attempt | On | On · Off |
| Flag a target that already has a live construct | On | On · Off |
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 portfolio of 14 constructs and 186 reports · “target” = the design goal, measured in the live solution · “estimated” = our estimate · research use; quoted facts link back to the original study record. People, companies and products named on this page are fictional — characters and sample data in the working solution.
A structured, cited register of every antibody–drug conjugate the team has built or inherited: target, antibody, linker, payload class, conjugation, drug-to-antibody ratio and distribution, aggregate and free payload, efficacy, tox findings and why each programme stopped. Scientists ask it questions in plain words and get answers that cite the page of the report each fact came from.
The prior-art reasoner compares the proposal with every construct on target, payload class, linker, conjugation and ratio, weighting precedents by how close the chemistry is. It returns a repeat-risk score with the nearest precedents and how they ended, each risk with its source, and the changes the portfolio shows work. The board chair decides.
From facts a named steward has confirmed, with a citation on every claim. Facts still waiting for a steward are marked and kept out of the conclusion, and below the confidence floor the answer says the evidence is thin and names what is missing.
Yes. The vocabulary mapper maps the other team's terms — “conjugate load”, “monomer loss”, “ocular surface finding” — onto your own and keeps the original wording on every field. When an inherited report and a transfer summary disagree, the conflict checker keeps both values on the record for a steward to decide.
No. Agents never resolve a conflict, never fill a field with no sentence behind it, and never round or convert a number silently. A fact becomes searchable only after a steward confirms it, and a prior-art check is accepted or sent back by the design board chair.
If you turn it on. Published papers sit beside your own reports and are always labelled, so an answer shows which facts are yours and which come from the literature.
Yes. Attaching a check to every proposal, flagging a payload class with a known organ finding, a target with a failed earlier attempt or a target that already has a live construct are settings — and a new rule, such as flagging a hydrophobic linker above a ratio of 6, can be added in plain words.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your study, CMC and tox reports, decision records and naming — and it goes live once every quality gate has passed. We will walk you through it on a sample of your own constructs first.
Ask the literature in plain words and get a sourced answer in seconds.
1-min filmPipeline WatchEvery rival trial, readout and approval lands in a sourced brief for each programme, every Monday.
1-min filmPreclinical ReportsRaw study data becomes a draft report in hours, with every number traced to its source.
1-min filmWe’ll run the ADC Knowledge Base on a sample of your own study, CMC and tox reports.