AI predicts
from sequence.Physis folds from evidence.
An all-atom glass-box refinement engine that turns sparse, noisy data and AI priors into physically-explainable structures — N < 110, in CPU-minutes, without ever exposing your IP.
AI predicts.
Physics proves.
PhysisFold is not another de-novo predictor. It is an Oracle Amplifier and an Explainable Refinement Layer — taking masked, noisy experimental data (NMR, mass-spec) or AI initial models, and applying the laws of physics to clean, rearrange and validate them into physically-consistent all-atom structures.
Sequence-to-structure models return a plausible fold. But for docking, FEP and synthesis, "plausible on paper" can still hide stereochemical errors and steric clashes that derail the work downstream.
A dynamic field of real physical forces relaxes and stabilizes the structure, resolving clashes and recovering native topology — and it explains every step it took to get there.
Four forces, one refinement engine.
Physics-consistent refinement
Folding is guided by a dynamic field of physical forces rather than statistical priors alone. The structure relaxes and stabilizes — avoiding the unnatural atomic contacts that statistics let slip through.
Oracle amplification
The Source-of-Information Oracle recovers structural signal under extreme degradation — up to 80% missing constraints — filling the gaps from thermodynamic equilibrium, never inventing detail the data can't support.
Explainable by design
No black box. Every run ships a detailed report — the physical forces that drove the conformation, a full value-data layer (secondary structure, networks, dynamics) and AlphaFold-style confidence & PAE — flagging any high-energy or unstable regions before they reach your bench.
Blind by construction
Runs server-side inside an AWS Nitro Enclave — no network, no persistent storage. Your sequence is end-to-end encrypted to the enclave (we only ever move ciphertext), and AWS-signed attestation lets you verify exactly what runs before you send a single byte. Use our managed AWS, or run the same enclave in your own account (BYOC).
Defined parameters.
Trustworthy results.
For reliability in the CADD community, PhysisFold operates inside clearly-stated bounds — so a PASS means what it says.
Peptides, cyclic peptides, de-novo designed binders, mini-proteins and single protein domains — exactly where evolutionary MSAs run dry.
Optimized for sub-110-residue chains, enabling fast high-throughput screening on conventional CPU resources.
Sparse NOEs from NMR, cross-linking mass-spec, template-derived distograms, or AI-generated priors — combined, not discarded.
Not just a PDB.
A verdict.
Every run delivers a complete, audit-ready package — the structure, an explainable JSON/Markdown report, a value-data layer single-structure predictors never produce, and a self-contained interactive HTML. All read-only, blind views of one physical fold, ending in a clear automated verdict.
Physics safety net
Feed it a bad template or a faulty restraint and it will not compromise into an unphysical structure. It reports the lj_severe clashes and forbidden Ramachandran angles instead — a built-in filter against false confidence.
Audit-ready reports
Structured JSON and Markdown logs trace which forces prevailed and climb an explainability ladder — from the worst residues, to the conflicting restraints behind them, to a blind keep-set for a cleaner second round and a non-gameable held-out check. Every run becomes a defensible record, easing regulatory compliance and IP documentation.
The signals you already trust
You still get a per-residue confidence (a pLDDT analog) and an N×N PAE map — the AlphaFold-style readouts your pipeline expects. In a deep ensemble it becomes genuinely calibrated: P(Cα error < 2 Å), validated on held-out data — never a restraint-fit proxy.
One fold, many views — structure, value-data layer, per-residue track, ensemble, confidence & PAE, and a self-contained interactive HTML. See everything a single run produces →
Flexible licensing.
Credit-based runs.
For the lab
Core engine and core physical forces for university research groups, on our managed enclave.
For the build
Commercial license, batch processing and full API/SDK access on our managed AWS enclave, for growing biotechs.
For scale
The same enclave deployed in your own AWS account, with custom integrations, OEM SDK and SLAs for pharma and CADD software vendors.
| Package | Total | Per run |
|---|---|---|
| 10 runs | €990 | €99.00 |
| 50 runs | €3,500 | €70.00 |
| 100 runs · most popular | €6,500 | €65.00 |
| 1,000 runs | €45,000 | €45.00 |
| Custom | Let's talk | unlimited / yr |
Validate every structure against physics before it drives a decision. Pay per fold, scale your credits with your pipeline, and keep a defensible, audit-ready record of every structure you ship.
Bring evidence.
Get a structure.
Give us a handful of distance restraints. We return a full bundle — an all-atom structure, an explainable quality report, AlphaFold-style confidence & PAE, and a self-contained interactive HTML — fast, blind by construction, and ready for downstream docking and synthesis.