This statement covers the Epitorix website and the products EPiQ™, Xenoclea, PharmPhax™, PharmacoLaw and the Epitorix AI models. It describes practice, not contract; the Terms of use and Privacy policy govern use.
Scientific integrity
Every figure we publish carries its method, its denominator and its stage in the same block as the number. Performance figures on this site come from retrospective analyses and internal benchmarks and are labelled as such where they appear; a preprint describing the evaluation is forthcoming. Where a metric was measured on a benchmark rather than in the clinic, we say so. A bare percentage is not a claim we make.
Human oversight
An output from our models is something for a scientist to review, not a decision. PharmPhax combines the models’ analysis with expert scientific review before a report reaches a client. EPiQ and Xenoclea return ranked designs with the predicted structures and data behind them, so the scientist can check the reasoning rather than accept a score. What to synthesize, test or advance stays a human decision.
Research use only
Epitorix is a research and development platform. Its predictions are not intended to independently diagnose, treat, prevent, or manage disease, and they should not be used as the sole basis for patient-level or clinical decisions.
The position is the same in our FAQ and our terms, and applies to every figure on this site. Outputs inform which candidates to advance and what to test next; they do not replace experimental validation, regulatory review or clinical judgment.
Provenance
EPiQ runs published engines alongside models built in house, and names both. RFpeptides, ColabDesign, BoltzGen, EvoDiff and the other published generators appear under their own names next to EpitorixRL, EpitorixDesignPeptide and EpitorixDesignPeptideSC on the EPiQ page. A published model keeps its name and is never passed off as our own. When we call a model ours, we mean an architecture designed by Epitorix, trained on data we generated and validated at the bench.
Evaluation and reporting
Our performance figures rest on two kinds of evidence, and each figure says which. Retrospective analyses test the models against what is already documented: radiopharmaceuticals with documented clinical outcomes, the basis for detecting previously reported development and safety liabilities with up to 93% accuracy, and known off-target interactions, the basis for over 90% recall in proteome-wide off-target prediction. Internal benchmarks are evaluations we ran ourselves; that is the basis for Pearson R 0.84 (peptides) and 0.70 (small molecules) between predicted and measured pKd. A preprint describing the evaluation is forthcoming. Until then, read these figures as internal results, not independently verified ones.
Data and intellectual property
As between Epitorix and the user, the user retains rights to the molecular outputs generated through their authorized EPiQ account, subject to the applicable platform agreement. Epitorix does not claim ownership of a user’s generated outputs solely because they were created using the platform. We do not guarantee that a generated molecule is unique: AI models may independently generate identical or substantially similar molecules for different users, particularly when they provide similar targets, binding sites, constraints or design objectives, and no output is warranted to be exclusive, novel or patentable. Patentability and freedom to operate should be assessed independently.
Responsible use
The platform is intended for legitimate therapeutic research. Requests for access are screened [[GAP: confirm policy — who screens, against which criteria]], and use that is inconsistent with that purpose is refused.
Privacy
Analytics on this site are cookieless. Cloudflare Web Analytics sets no cookies, stores nothing in your browser, and strips query strings and credentials from the page addresses it records; no consent banner is shown because no tracking cookie is set. Inputs you submit to the products — targets, sequences, constraints, results — stay within your account and are not used to train models offered to other customers [[GAP: confirm]]. How we handle personal information is described in the Privacy policy.
Environmental footprint
Training and running models costs energy. [[GAP: environmental footprint — where Epitorix compute runs, how its energy use is measured, and what Epitorix commits to report. No source supports a figure or a commitment; publish this section only once the founder supplies one.]]
Our limits
Plainly, what the models do not do. They make no clinical guarantees. A predicted Kd, an off-target hit or an organ-risk flag is an enrichment filter: it changes which candidates are worth testing, not what happens in a patient. They do not replace the experiments that follow, or the regulatory path before any trial. Every result on this site is preclinical: our lead candidates are being tested in vivo, and clinical trials remain ahead.
Reporting a concern
If a figure on this site looks wrong, an output raises a safety concern, or you believe the platform is being misused, tell us at support@epitorix.com or through the contact form. We acknowledge every report. Where a published figure is shown to be wrong, we correct the page and say what changed [[GAP: confirm corrections practice]].