Discovery finds the target. Safety models the body. What happens in between?
Innophore’s Catalophore™ point-cloud technology is the layer that maps structure-based off-target binding across the entire proteome — connecting molecular discovery with safety and DMPK modeling. Patented, GPU-accelerated, and validated with a global pharma leader.
A blind spot between optimizing a molecule and trusting it.
Discovery tools refine a ligand against a known target. Safety models predict what the body does with it. Neither answers the question in the middle — and roughly 30% of drugs fail clinical trials over safety concerns, with off-target toxicity among the leading causes.
On-target discovery
Ligands are designed and scored against the intended pocket. Docking, free energy, structure — focused on a single, known target.
Off-target binding
Unrelated proteins can evolve convergently similar pockets. A molecule may bind targets sharing no sequence or fold — invisible to sequence-based screening.
Safety & DMPK
Standard toxicity panels typically cover fewer than 60 protein targets — a fraction of the possible off-target space. PBPK and PK/PD models then predict risk, but only for the targets they are given.
Two proteins that share no evolutionary ancestry can still develop pockets shaped alike. Sequence-based methods never see it — cavity similarity does.
The convergent evolution principleA horizontal layer for proteome-wide cavity analysis.
The Catalophore™ technology describes binding pockets as 3D point-cloud signatures of their physico-chemical property fields, then searches the entire structural proteome for functionally similar cavities — independent of sequence, fold or family. Built to sit above any discovery stack and feed any safety model. Not an island. A layer.
A cavity layer is only as trustworthy as the structures beneath it. AI structure prediction is transformative — but it still gets the chemistry wrong often enough to matter: by one benchmark, chirality errors occur in 4.4% of cases. At proteome scale, that is thousands of subtly wrong binding sites.
That is why every structure entering the cavitome is energy-minimized, refined, and curated — not taken on faith. Off-target prediction is a safety question; the inputs have to be right.
From a lead molecule to a mechanistic safety signal.
One continuous path. Each stage hands the next exactly what it needs.
Discovery platform
Delivers an optimized ligand and its validated binding mode against the intended target.
Innophore
Compares the binding site against 849,355 cavity descriptors covering the human proteome — ligand-agnostic. Returns a ranked off-target list, each with a structural rationale.
Safety & DMPK modeling
Translates the identified targets into physiological consequence — exposure, margins, interaction risk.
Proven across discovery programs.
The Cavity Layer is not a concept — it has been applied and published in peer-reviewed work.
Off-target triage
Screen a binding site proteome-wide before lead nomination — ligand-agnostic, so it works before a compound is even fully characterized.
Joint study with a global pharmaceutical company, 2026
Mechanistic rationale
Engineer antivirals with a broader spectrum and fewer off-target liabilities — giving a safety signal a structural cause.
Viruses 16, 1186 (2024) ↗
Target deconvolution
Resolve binding sites proteome-wide across full viral genomes — from monkeypox to SARS-CoV-2 target proteins.
Microbiology Spectrum 11 (2023) ↗
Beyond pharma
The same cavity search discovers new biocatalysts — proof the layer generalizes across all of protein space.
ChemBioChem 25 (2024) ↗
New uses for known drugs
The same cavity search that flags off-targets also finds repurposing opportunities — approved drugs whose binding sites match a new target.
Scientific Reports 13, 11783 (2023) ↗
The Catalophore™ origin
The proof-of-principle: mining structural databases by active-site constellation to find promiscuous activity.
Nature Communications 5 (2014) ↗
Patented, benchmarked, validated with global pharma.
The Cavity Layer is not experimental. Its core method is a granted patent, accelerated on production hardware, and validated for early toxicity screening in a joint study with a major global pharmaceutical company.
Protected core method
The point-cloud cavity method — representing protein pockets by their physico-chemical property fields — is a granted patent. US 2015/0302142 A1 ↗ · also granted as US 10,825,547 B2 and across 15 jurisdictions.
100× on production GPUs
On enterprise GPU systems, cavity-matching throughput was accelerated by a factor exceeding 100 — a proteome match that took ~625 seconds now completes in roughly 5.
Official NVIDIA partner
Innophore is a listed NVIDIA healthcare & life sciences partner. The proteome-wide structural dataset behind the cavitome was built jointly with NVIDIA on BioNeMo — generated in two weeks, a task that previously took over a year.
Validated with a global pharma leader
Co-developed and validated in a joint study with a major global pharmaceutical company, across 467 experimentally confirmed drug–target pairs.
Validated against 467 experimentally confirmed drug–target pairs, the method recovers off-targets across structurally unrelated protein families — operating without any ligand information, sequence, or overall protein structure.
Ligand-agnostic off-target validation“NVIDIA boosted our performance so that we can run five million off-target predictions per second. We’ve been at a few hundred before.”
Christian C. Gruber · CEO, Innophore
Ligand-agnostic off-target site prediction for early toxicity screening by leveraging point cloud-based protein cavity analysis
Joint study with a major global pharmaceutical company — cavity-based off-target validation across 467 drug–target pairs
Folding the human proteome using BioNeMo: a fused dataset of structural models for machine learning
Nature Scientific Data 11, 591 — proteome-scale structural foundation, built with NVIDIA
CavitOmiX Drug Discovery: engineering antivirals with enhanced spectrum and reduced side effects
Viruses 16, 1186 — off-target reduction by cavity design
Fusidic & flufenamic acid identified as SARS-CoV-2 inhibitors via DrugSolver CavitomiX
Scientific Reports 13, 11783 — drug repurposing by cavity similarity
Identification of promiscuous ene-reductase activity by mining structural databases using active site constellations
Nature Communications 5 — the founding Catalophore™ proof-of-principle
Selected publications. Partner reference shown in generic form pending naming clearance.
Plugs into the stack you already run.
The Cavity Layer is a product, not a project. Standard interfaces, defined inputs and outputs, no dependency on any single vendor.
- API
Standard REST API
Programmatic submission of structures, retrieval of ranked cavity matches.
- I/O
Model-ready outputs
Off-target lists formatted to drop straight into downstream safety pipelines.
- ∞
Platform-agnostic
No lock-in — connects to discovery and safety environments regardless of vendor.
# submit a ligand + binding pose POST /v1/cavity/scan { "structure": "<pdb>", "pose": "<ligand>", "scope": "proteome" } # → ranked off-targets + rationale { "matches": [ { "target": "…", "similarity": 0.91, "evidence": "cavity" } ] }
Put the layer to work.
Whether you run discovery programs or build the platforms behind them — there’s a way in.
Request a pilot
Bring a lead series. We run a proteome-wide off-target scan and walk you through the structural evidence.
Start a pilot →Explore integration
Add proteome-wide cavity analysis to your discovery or safety environment via a standard API.
Talk integration →