01 / Gnosis layer
Gnosis I, Target Binding
Predicts binding affinity across 9 oncology targets. Every top candidate is auto-docked into the 3D pocket of the relevant protein.
9 targets · auto 3D docking · pocket-resolved
Multi-modular integrated AI drug discovery ecosystem
Six AI modules. One drug-discovery decision.
Every candidate scored on six dimensions: binding, selectivity, activity, safety, gene expression, ADMET. Ranked, docked shortlist returned in minutes.
Six AI modules. One integrated decision layer.
01 / 06
02 / Gnosis layer
Gnosis II, Cancer vs Normal Cell Selectivity
Predicts IC50 and activity across cancer and normal cell lines. A cancer-vs-normal selectivity index downranks toxic non-selective compounds automatically.
700+ cancer cell lines · normal-cell selectivity index
03 / Gnosis layer
Gnosis III, Clinical Safety
Nine preclinical toxicity endpoints plus six organ-level clinical AE categories. Safety gates suppress favourable predictions when toxicity flags fire. See it run live on doxorubicin.
9 endpoints · multi-organ panel
04 / Gnosis layer
Gnosis IV, Downstream Omics
Predicts the differentially expressed genes and Hallmark pathways a compound triggers, per cell line, dose, and timepoint. Drives MoA and resistance strategy.
946 landmark genes · 49 Hallmark pathways · 76 cell-line contexts
05 / Gnosis layer
Gnosis V, Drug Behaviour (ADMET)
Twelve ADMET models covering absorption, distribution, metabolism, excretion, and toxicity.
12 models · 5-enzyme CYP panel
06 / Gnosis layer
Gnosis Forge, Generation Engine
Every molecule Forge generates is scored in real time by all five Gnosis modules simultaneously. Weak dimensions become correction signals mid-generation. The result: candidates that are not just potent, but carry the right safety profile, metabolic behaviour, and genomic fingerprint from the first round.
50+ candidates per campaign · 142 mechanism categories
02 / the problem
90% of drug candidates fail in clinical trials. Most fail because they're evaluated one dimension at a time.
Clinical failure rates remain near 90% and time-to-patient has barely moved in three decades. Not because AI tools are missing, but because candidates are still evaluated one dimension at a time. Fatal flaws surface after hundreds of millions have been spent.
Vecentra evaluates every dimension at once. Fatal flaws surface on day one, not after $300M in spend.
$2.6B
Average cost per approved drug, including failures. (Deloitte, 2023)
~90%
Clinical trial failure rate.
10 to 12 yr
From target discovery to first patient dose in conventional workflows.
~50%
Of late-stage clinical failures driven by safety, not efficacy.
03 / positioning
Fragmented AI produces fragmented decisions. Vecentra integrates every signal.
Not one algorithm, a multi-modular ecosystem. Six AI modules covering binding, selectivity, activity, safety, gene expression and ADMET, each informing the others and reconciled by a contradiction engine that catches disagreements before a researcher sees them. The same signals steer Gnosis Forge, so new molecules are built potent, selective, safe and metabolically sound from the first round.
One input. One platform. A complete picture.
04 / scope
Two indications, one platform.
GNOSIS is built as a disease-agnostic architecture. Oncology is the lead program, with metastasis as the entry wedge. Parkinson's Disease is the second pillar, a market with almost no equivalent integrated AI platforms.
05 / where vecentra fits
The integration gap is where Vecentra sits.
The AI drug discovery field has produced excellent specialists. Each solves one hard problem well. A team that needs the full picture must assemble five separate tools, five separate contracts, and reconcile the outputs manually.
| Capability | Vecentra | Atomwise | Insilico | Recursion | BenevolentAI | Schrödinger |
|---|---|---|---|---|---|---|
| Target binding | ✓ | ✓ (needs 3D) | ✓ | - | ✓ | ✓ |
| Cancer-vs-normal cell selectivity | ✓ | - | - | partial | - | - |
| Cancer-cell cytotoxicity | ✓ | - | - | partial | - | - |
| Clinical safety (multi-endpoint) | ✓ | - | partial | - | partial | partial |
| Gene expression (946 genes) | ✓ | - | - | partial | - | - |
| ADMET (12 models) | ✓ | - | partial | - | - | partial |
| AI molecular generation | ✓ | - | ✓ | ✓ | - | partial |
| Integrated 3D docking | ✓ | partial | - | - | - | ✓ |
| Cross-module validation | ✓ | - | - | - | - | - |
| Single SMILES input | ✓ | - | - | - | - | - |
| No protein structure required | ✓ | - | ✓ | ✓ | ✓ | - |
✓ present in platform · partial = limited scope · - = not a focus of this platform
Vecentra is not competing with Schrödinger on physics-based binding accuracy or with Recursion on imaging scale. Vecentra is the platform that connects every evaluation dimension into one coherent ranking, from a single SMILES input, with selectivity built in, in one workflow.
06 / evidence
Most AI tools show you their wins. We publish the hard half.
A number you cannot trust is worse than no number. Every result below carries the limit that qualifies it, including the compounds we get wrong and the targets where a naive approach fails.
Safety
It caught the liver failures before the clinic
On an external panel of real drugs, it cleared 7 of 8 approved medicines and flagged all 3 later pulled from development for liver injury, the late-stage surprise you most want to catch early.
Read the method noteBinding
Binding ranked on par with physics, in minutes
On public free-energy benchmarks it ranks binding in the same band as the physics methods medicinal chemists trust, minutes per compound, not days, and holds up on blinded targets where naive scoring collapses.
See the benchmarksDiscipline
It tells you when not to trust it
Every prediction carries its uncertainty and flags molecules outside its proven chemistry, so the calls you act on are ones the platform stands behind, and borderline cases go to assay, not into a false positive.
0.000
hERG risk · AUROC
0.000
CYP1A2 · AUROC
halved
solubility error
0.000
micro-stability · beats ref
recovered
binding rank
Measured on held-out public benchmarks. Full method and limits on the benchmarks page.
07 / no black box
Every number comes with its reasoning.
Gnosis Lens opens each prediction. It shows the exact part of a molecule that drove a safety or ADMET call, names the structural feature behind it, and flags where the model is uncertain. You check the reasoning against known chemistry, not a score you take on faith.
On-structure attribution
The atoms that drove each call, mapped on the 2D structure.
Named drivers
The functional group behind every liability, verdict-coloured.
Stated uncertainty
Out-of-distribution and low-confidence calls are flagged, not hidden.
lowers flagattributionraises flag
Nitroaromatic groupraises
Aromatic amineraises
High polar surface arealowers
sp3-rich scaffoldlowers
Illustrative. Gnosis Lens attributes every live prediction to the structure that drove it, coloured by verdict and checked against known chemistry.
08 / how partners use it
Three ways partners use the ecosystem today.
Evaluate a known compound
Score one of your in-pipeline leads across all six dimensions: binding, selectivity, cytotoxicity, safety, gene expression, ADMET, plus an automatic 3D dock. Use to triage, optimise, or de-risk before screening spend.
Design from scratch
Give Vecentra a target, disease, or mechanism objective. Forge returns 50+ ranked, scored, docked candidates with built-in selectivity constraints, no parent compound or protein structure required.
Profile a competitor or failed asset
Find out why a clinical-stage compound failed, or assess a published competitor structure before committing to your own program. Often the answer is in the selectivity profile.
09 / on the roadmap
The platform is live today. Three things are coming next.
Feature-level steering that pushes generation away from liabilities as it designs. Specialised modules for hard biology beyond oncology. Autonomous discovery agents that close the loop from analysis to candidate without human triggering.
Detail and timelines available under NDA.
10 / frequently asked
Questions we get from partners and investors.
11 / start now
For research and BD partners
Bring Vecentra into the loop before synthesis, screening, or program lock-in. We will run a representative campaign on a compound of your choice within an agreed timeframe.
Discuss a programFor investors
Series A materials, platform data, and a 30-minute platform walkthrough available on request.
Request investor materials