The failures-first diligence layer for biotech.
We combine public clinical data with proprietary trial intelligence to unlock hidden insights from hard-to-find negative outcomes.
Find what didn’t work, before you fund what won’t.
Natural language search
Ask about a drug, target, mechanism, or trial.
Drag & drop PDFs, protocols, or CSRs (optional)
Grids
| Company | Drug / Candidate | Indication | MoA | Phase / Design | Key issue |
|---|---|---|---|---|---|
| Quince Therapeutics | eDSP (EryDex) | A-T | DSP | Ph 3 RCT | Endpoint miss |
| Biohaven / AZ | Verdiperstat | ALS | MPO | Ph 3 RCT | No efficacy |
| Applied Therapeutics | Govorestat | Galactosemia | AR | Ph 3 | Design / endpoint risk |
Failed clinical trials data
Quince Therapeutics — eDSP (EryDex)
3 sources found
How it works
Failures-first diligence in three steps.
We ingest target data
Mechanisms, preclinical data, protocols, readouts, and technical analysis.
We find precedents, including hard-to-find failures
We match by target, modality, indication, endpoint, population, and design using public clinical records plus proprietary trial intelligence.
You receive hidden insights from buried data
Comparable programs, what’s similar vs different, failure modes, salvageability, and a source pack tagged by provenance.
Every claim is traceable. Every source is labeled (public, rare, proprietary).
Our platform
Four layers of failures-first intelligence, from question to provenance.
Natural language search
- Ask if an approach has failed before, and why.
- Query by target, mechanism, drug, indication, endpoint, or trial.
- Returns closest precedents, what’s similar, what’s different.
Grids
- Standardize diligence into a table you can trust.
- Choose Failure Analysis or Scientific Diligence.
- Auto-populates comparable programs and key trial attributes.
Failure analysis
- Identify failure modes and repeat-risk.
- Endpoint miss vs biology vs execution, made explicit.
- Outputs salvageability and what must change.
Source provenance
- Every claim tagged by where it came from.
- Public, rare, and proprietary sources, clearly labeled.
- Translations and deep-search flags where relevant.
Natural language search
Ask about a drug, target, mechanism, or trial.
Grids
| Company | Drug | Indication | MoA | Phase | Key issue |
|---|---|---|---|---|---|
| Quince Therapeutics | eDSP (EryDex) | A-T | DSP | Ph 3 RCT | Endpoint miss |
| Biohaven / AZ | Verdiperstat | ALS | MPO | Ph 3 RCT | No efficacy |
| Applied Therapeutics | Govorestat | Galactosemia | AR | Ph 3 | Design risk |
Failure Analysis
| Company | Drug | Key issue | Reason for failure | Salvageable? |
|---|---|---|---|---|
| Quince Therapeutics | eDSP | Endpoint miss | Underpowered for rare disease | Possible |
| Biohaven / AZ | Verdiperstat | No efficacy | Target not validated in ALS | Unlikely |
| Applied Therapeutics | Govorestat | Design risk | Endpoint not clinically meaningful | Possible |
Source provenance
Quince Therapeutics — eDSP (EryDex)
3 sources foundThe evidence base is incomplete. That bias costs money.
Scientific diligence is systematically skewed toward positive narratives. Disappointing results are often unpublished, fragmented, or buried in non-standard disclosures, so teams underwrite programs with a structurally incomplete map of what has already failed.
Without a failures-first view, investors and pharma BD repeatedly fund approaches that have already broken for the same underlying reasons: endpoint mismatch, insufficient target engagement, tolerability limits, or weak translational bridges.
The result is wasted R&D spend and slower progress for patients. Onyx is built to make negative outcomes legible, comparable, and citable, so capital decisions reflect the full evidence base, not just what gets published.
Sources: The Lancet (publication gap); Nature Drug Discovery (waste estimates).
Why Onyx
Enterprise tools are strong on published pipelines and filings. Generic LLMs inherit the internet’s success bias. Onyx is built for failures-first diligence.
| Capability | Onyx | Enterprise platforms | Generic LLMs |
|---|---|---|---|
| Hard-to-find negative outcomes | ✅ | ⚠️ | ⚠️ |
| Dataroom ingestion | ✅ | ⚠️ | — |
| Similarity matching (MoA/modality/endpoint/population) | ✅ | ⚠️ | ⚠️ |
| What’s similar vs different | ✅ | ⚠️ | ⚠️ |
| Source provenance (public/rare/proprietary) | ✅ | — | — |
| Standardized grid + dossier | ✅ | ✅ | ⚠️ |
| Auditable citations/links | ✅ | ✅ | ⚠️ |
⚠️ indicates partial coverage or inconsistent availability.
FAQ
What do you mean by ‘proprietary trial intelligence’?
We compile hard-to-find trial signals from our biotech network and non-standard disclosures, then tag every claim by provenance. We don’t present unattributed assertions.
Do you use confidential client data to train models?
No. Client materials stay private and are used only to generate outputs for that workspace.
How do you avoid hallucinations?
We ground outputs in citable sources, separate facts from interpretation, and flag uncertainty. If we can’t substantiate a claim, we say so.
What sources do you pull from?
Public clinical registries, patents, abstracts and disclosures, regulatory documents, plus Onyx proprietary intelligence where available, all clearly labeled.
Is this investment advice?
No. We provide scientific diligence and scenario framing, not price targets or recommendations.
Who is this for?
Biotech investors, pharma BD, and operators doing diligence on programs, platforms, and pipelines.
How does access work?
We’re running a private beta. Request access for a demo and coverage discussion.
Stop funding the same failure twice.
Onyx turns scattered negative outcomes into a failures-first map you can underwrite. We ingest a target’s data room, surface precedents and failure modes, and show what would need to change to work.
Built for biotech investors and BD teams. Evidence is cited, sources are labeled.