R&D integration after an acquisition: every inherited file read, every fact cited, a signed dossier for each asset
Every inherited file read and cited, and a signed dossier for each asset before the experts leave.
See one case, screen by screen ↓The deal closes and the clock starts. The integration lead inherits tens of thousands of files — the deal data room, a lab notebook export, LIMS result sets, batch and stability records, CRO study reports and the agency correspondence — and a 100-day plan that expects a signed dossier for every asset. Meanwhile the head of DMPK, the VP of pharmacology and the process chemistry lead all have leaving dates.
The hard part is not the reading. It is knowing which version counts: the IND summary that quotes a draft NOAEL, the CMC report that never heard about the polymorph a chemist found in the notebook, the toxicokinetic raw data that never reached the data room. By hand, the function heads find these one at a time — often after the person who could have explained them has gone, and sometimes after a new animal study has been ordered that the acquired company had already run.
Estimated split for one acquired asset’s dossier, by hand and with the solution.
Six specialist agents classify, extract, cross-check, draft and answer “has this been done before?”; each function owner resolves and signs.
Day 50 of the Halden Therapeutics integration. HLD-217, an oral NLRP3 inhibitor, has finished its Phase 1 single ascending dose study and its IND has just been transferred. Here is what happens next, screen by screen, in the working solution.
Priya opens the solution on day 50 of the 100-day plan: 38,412 inherited files, 84% classified, and four assets with their dossier readiness. Beside the file map is the list that sets the pace — Dr. Ingrid Solheim, Head of DMPK, leaves in 24 days; Dr. Elias Brandt, VP Pharmacology, in 38.
Needs you: “Dog NOAEL: the IND summary and the final report disagree — HLD-217 toxicology · 50 vs 15 mg/kg/day · Dr. Maya Chen decides.”
214 files (1.2 GB) arrived overnight on the Larkfield Preclinical share. The files are read where they are; nothing is copied or changed. The classifier sorts them into 41 study reports, 96 data tables, 58 letters and minutes and 19 duplicates, keeps the GLP status on all 41 reports and marks two drafts as drafts. The extractor pulls 612 facts, each with its page, and adds three new studies to the registry.
“Halden had already drafted the 13-week dog protocol and the QTc analysis plan — two commitments are further along than we knew.”
In the IND 158204 toxicology written summary (Module 2.6.6), the agent highlights one sentence: the dog NOAEL was 50 mg/kg/day — taken from the draft report of November 2, 2023. The final report, Amendment 1, says 15. The file is marked “Sources disagree”, and its original status, submitted to the agency, is kept.
The file is read where it lives; the dossier cites it and never changes it.
Maya opens the HLD-217 toxicology section. The agents show how they built it: 1,412 files labelled for toxicology with drafts and finals told apart, five key facts with their pages, one contradiction, two gaps, eight citations. The contradiction sits on top — the IND value and the final report value side by side, with the conclusion page of the final report open on the right.
Recommendation: “Use the final report. Amendment 1 supersedes the draft, and the signed pathology peer review supports it.”
Maya chooses the final report — 15 mg/kg/day. The dossier is updated in three places, and a new gap assigns the correction of the IND summary to Dana Okafor for the next annual report. On the final value the exposure margin to the planned 200 mg Phase 2 dose is 1.9×, not 6.2×. Keeping the IND value instead would have needed a written reason, recorded with the decision.
With no contradiction open, the section can be signed. The sign-off shows the section, five key facts each cited to its page, and the signer. Maya picks the meaning of her signature — “I reviewed the cited evidence and approve this section for the integration dossier” — and enters her password.
While a contradiction is still open, the signature button stays disabled.
HLD-217 moves to 78% ready, three of six sections signed. The draft section reads plainly: the dog NOAEL is 15 mg/kg/day from the final report, supported by the pathology peer review, and trials dosing longer than 28 days need 3-month studies in two species first. Each sentence carries its citation, and the gaps for the section sit underneath with their status.
Human metabolite M3 is 14.2% of drug-related exposure and was not quantified in rat plasma — and the person who would know is Dr. Ingrid Solheim. Her question pack is ready: four short questions, each linked to its gap and its source, such as whether M3 was measured in any rat or dog plasma sample, even outside the GLP studies, and where the raw toxicokinetic files for rat study TX-004 are.
Answers are recorded as interviews linked to the gaps — never as source records.
Omar Haddad, program leader, is planning Phase 2 dose selection. The advisor lists 20 candidate mouse models, searches 412 study reports and 9,742 notebook entries, and finds 7 that Halden ran or tried, 3 full studies and 2 to reuse: MSU crystal peritonitis (IL-1β down 72% at 30 mg/kg) and the LPS/ATP challenge (ED50 3.1 mg/kg, matching human whole-blood potency). Only a lower 3 mg/kg arm is still needed.
The CAPS knock-in study is shown too — stopped at week 3, colony unstable.
Files classified since the data room arrived on day 14, against the day-30 target of 75%; facts extracted by function; dossier readiness by asset and function; what the record is made of; and open gaps by owner with the next due date — the view Priya takes to the integration review.
The same acquisition, seen by the people who carry it — the integration lead, the function owners who sign, and the program leader planning what comes next.
Six specialist agents classify, extract, cross-check, draft and answer “has this been done before?”; each function owner resolves and signs.
Reads every inherited file in place and labels its record type, asset, study, version and original GLP or GMP status; finds duplicates and drafts.
Pulls studies, findings, PK parameters, lots, stability results and agency commitments into the science registry — each with its document, version and page.
Compares the same fact across versions and sources, checks the package against agency commitments and guidance, and raises gaps with an owner and an expert who could answer them.
Drafts each asset dossier section — toxicology, DMPK, CMC, regulatory, IP, clinical — every sentence cited, open questions stated plainly.
Builds a question pack for each departing scientist from the open gaps they authored or know about, and links their answers back to the gaps.
Answers “has this been done before?” — matches a planned study against the acquired company’s reports and notebooks and recommends what to reuse.
Resolves conflicts and signs. The agents propose; a named person decides.
Studies, lots, commitments, people — ask in plain words, or tell it what to change. Every answer is cited.
What contradicts the HLD-217 dog NOAEL?
Two sources disagree. The IND toxicology summary says 50 mg/kg/day, taken from the draft report. The final report, Amendment 1, says 15 mg/kg/day, after a pathology peer review found minimal liver necrosis at 50. On the final value the margin to the planned 200 mg Phase 2 dose is 1.9× instead of 6.2×.
Has Halden already run an NLRP3 mouse model?
Yes — two with clear results. MSU crystal peritonitis cut IL-1β by 72% at 30 mg/kg, and the LPS/ATP challenge gave an ED50 of 3.1 mg/kg that matches human whole-blood potency. A CAPS knock-in study was stopped at week 3.
Who leaves next, and what do they know?
Dr. Ingrid Solheim leaves Oct 31, in 24 days, with gaps on metabolite M3 and the rat raw data; Dr. Elias Brandt leaves Nov 14, on the mouse models; Marco Velasquez leaves Dec 19, on Form B and the conjugate lots. Dr. Solheim is first, and her 4-question pack is ready.
Add a rule: flag human metabolites above 10% without tox-species coverage
Done. Every metabolite report is now checked against rat and dog exposure, and an uncovered metabolite above 10% raises a DMPK gap — as with M3 at 14.2%, which the agency asked Halden to evaluate. The rule is on in Settings and recorded in the audit trail.
13 screens from the working solution, on its sample data. Pick one to see it large.
Who signs, what the agents may do and when to start talking to the people leaving — all settings.
| Setting | Default | Choose from |
|---|---|---|
| Who signs each dossier section | Its function owner | Any named person, per section |
| Cite the source page for every fact | Always on | Cannot be turned off |
| Keep the original GLP and GMP status | Always on | Cannot be turned off |
| Prefer final and amended reports over drafts | On | On · off |
| Flag values that differ between sources | On | On · off |
| Send a fact to a person below confidence | 0.85 | 0.80 · 0.85 · 0.90 |
| Remind me before a scientist leaves | 30 days before | 14 · 30 · 45 days |
| Build question packs automatically | On | When a gap names someone who is leaving |
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 integration data · “target” = the design goal, measured in the live solution · “estimated” = our estimate, including the by-hand baseline · “typical” = published figures (a published vendor pilot at a top-20 drugmaker: 20 candidate models narrowed to 2, about 8 months of animal work saved). People, companies, assets and studies named on this page are characters in the working solution.
A solution for R&D integration after an acquisition. Six agents read the acquired company’s data room, lab notebooks, LIMS results, batch and stability records, CRO reports and agency correspondence, pull the facts into a searchable science registry with their pages, flag contradictions and gaps, and draft a cited dossier for each asset that the function owners sign.
The Contradiction & Gap Finder compares the same fact across versions and sources, shows both values with their pages and recommends which one supersedes the other. Only the section owner decides; once decided, the Dossier Writer rewrites every sentence that used the value, and keeping a superseded value needs a recorded reason.
No. Files are read where they live and never copied or changed, and the original GLP or GMP status is kept. The dossier cites the records; it does not replace them. Both rules are always on.
It tracks each person’s leaving date and builds a question pack from the open gaps linked to files they authored or studies they ran — short questions a yes, a number or a location can answer. Answers are recorded as interviews linked to the gaps, and if an answer names a file, the extractor goes to find it.
Yes. The Study Reuse Advisor reads the planned study, lists candidate models, searches the acquired company’s study reports and notebook entries, and recommends what to reuse and what is still needed — citing every prior result and saying when a study was stopped or inconclusive.
The deal data room export, the lab notebook export, LIMS results, batch and stability records, decks, SOPs and minutes, CRO study reports and regulatory correspondence — all read only.
Yes. The agents classify, extract, cross-check and draft; a named function owner resolves every contradiction and signs every section, with the meaning of the signature recorded. Each agent step and each human decision is on the audit trail.
The Agentic Solution Engine builds and deploys it from your requirements — your dossier templates, sign-off owners, record types and a sample of an inherited data room — and it goes live once every quality gate has passed. We will walk you through it on your own files 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 Science Harvester on a sample of an inherited data room of your own.