Live discovery console • simulated data

Somewhere in 10⁶⁰ molecules is the one that works.

PROTEUS is Ace Hacker's quantum-accelerated discovery engine. It predicts how a target protein folds, generates novel molecules to fit it, computes their binding energy with quantum chemistry, and screens for safety, collapsing years of wet-lab iteration into compute your chemists can steer.

Request research access → Explore the engine

1.2 kcal/molbinding-energy MAE

10⁸/daycandidates screened

40×fewer wet-lab cycles

proteus://bind/live-simulation CONVERGING

ligand · target complex

binding energy ΔG−9.4 kcal/mol

Candidates / sec

1,284

Best affinity

−11.2pKd 8.2

Ansatz depth

14UCCSD

Engine at a glance

Compounds evaluated in silico per pilot

Screen

0 kcal/mol

Binding-energy mean absolute error

Bind / VQE

Pocket prediction accuracy (held-out)

Fold

Fewer synthesis cycles to a validated lead

End to end

One engine, four instruments

The whole discovery loop, from target to lead.

PROTEUS pairs each step with the right computation, deep learning where pattern beats first-principles, quantum chemistry where electrons must be simulated honestly, and keeps a chemist in the loop at every gate.

Fold

Structure & pockets

Predicts the target's 3D fold and finds the druggable pockets, with per-residue confidence.

structure model • contact maps

Forge

Generative design

Invents novel, synthesizable molecules shaped to the pocket and to your property targets.

diffusion • RL • retro-synthesis

Bind

Quantum chemistry

Computes binding energy with variational quantum eigensolvers where classical DFT falls short.

VQE • UCCSD • embedding

Screen

Screening & ADMET

Filters millions of candidates for potency, safety, and developability before anyone lifts a pipette.

ADMET • tox • selectivity

Fold • Instrument 01

Before you drug a target, you have to see it.

Fold predicts the target protein's three-dimensional structure from sequence, maps which residues touch which, and locates the pockets a molecule could actually bind, each annotated with a confidence you can trust or challenge.

01
Structure from sequenceA folding model predicts the 3D backbone and side chains, with per-residue confidence (pLDDT) so you know which regions to trust.
02
Contact & distance mapsThe residue-residue contact map exposes the fold's topology and the interfaces that matter for binding.
03
Pocket detectionDruggable cavities are found and ranked by volume, hydrophobicity, and conservation, the sites Forge will design against.

Evoformer-styleMSA featurespLDDTfpocket

Residue contact map148 residues

residue i × residue j

Contact (< 8Å)Predicted interfaceNo contact

Per-residue confidence (pLDDT)mean 91.4

Binding pocket detectiontop 3 ranked

volume × hydrophobicity

Forge • Instrument 02

Do not search the catalog. Invent the molecule.

Forge generates novel chemical matter conditioned on the pocket and on your property targets at once. It explores the parts of chemical space no vendor library contains, and only proposes molecules a chemist could actually make.

Chemical space • generated library4,800 molecules

latent projection (UMAP)

GeneratedPareto-optimalSelected for synthesis

01
Pocket-conditioned diffusionA diffusion model grows molecules directly inside the binding pocket, so geometry is right by construction.
02
Multi-objective by designPotency, selectivity, solubility, and synthetic accessibility are optimized together, not traded off after the fact.
03
Synthesizable onlyA retro-synthesis check gates every candidate, so what Forge proposes, your medicinal chemists can make.

Equivariant diffusionRL fine-tuningQED / SA scoreretro-synthesis

Drug-likeness profilelead candidate vs target

Diversity & potency over generationsRL fine-tuning

Bind • Instrument 03

Binding is a quantum event. We simulate it as one.

Whether a molecule binds comes down to electrons, and electron correlation is exactly where classical density-functional theory makes its largest errors. Bind uses variational quantum eigensolvers on the active region to compute binding energies with chemical accuracy as the hardware matures.

VQE energy convergenceground-state energy

iteration → ⟨H⟩ (Hartree)

VQE (UCCSD)Classical DFTReference (FCI)

Potential energy surfacedihedral scan

Computed binding affinity • ranked candidatesΔG kcal/mol • lower is stronger

Screen • Instrument 04

Kill the bad molecules before the bench does.

Most candidates fail on safety and developability, not potency. Screen predicts absorption, distribution, metabolism, excretion, and toxicity across dozens of endpoints, narrowing millions of molecules to a handful worth synthesizing.

01
Cascade of filtersPotency, then selectivity, then ADMET, then developability, each stage cheap enough to run on everything upstream.
02
Toxicophore alertsKnown liability substructures and predicted off-target hits are flagged with the evidence behind them.
03
Calibrated, not black-boxEvery prediction carries an uncertainty, so a borderline call goes to a human, not silently through.

hERGCYP450AMESclearancesolubility

Discovery funnelper campaign

ADMET risk map • candidate × endpointrisk 0–1

Predicted potency distributionscreened library

hit cutoff —

Benchmarks • honest numbers

Where quantum earns its place, and where it does not.

We benchmark against strong classical baselines on public and internal sets. Quantum chemistry wins on the correlated, small active regions that decide binding; deep learning wins on throughput. PROTEUS routes accordingly.

Binding-energy error vs methodMAE kcal/mol • lower better

Baseline methodchemical-accuracy line

Capability profilevs incumbent CADD stack

Pipeline • from target to lead

One loop, closed by data and by chemists.

Each instrument feeds the next, and every gate is a place a human decides. Assay results flow back to retrain the models, so the engine gets sharper with each campaign.

Now partnering on programs

Bring us a target. We’ll bring the molecules.

PROTEUS is in research partnerships with a small number of teams. If you have a validated target and a hard chemistry problem, we should design against it together.

Request research access → Read the methods paper

Secure enclave deployment • your IP stays yours • wet-lab validation partners on request