Ranked candidate shortlist
Ordered hypotheses with a match value so attention starts on the structures the data can actually support.
Compound identification
A scientific agent that combines large-scale retrieval, foundation models and physics-guided validation to verify knowns, dereplicate mixtures and identify unknowns.
Built for chemists, researchers and lab teams.
01
Bring raw 1H and 13C data from Bruker, JEOL, Varian or JCAMP-DX.
02
Search large-scale compound libraries and rank plausible structures with foundation models.
03
Compare predicted and measured shifts to discard candidates the data cannot support.
04
Follow the reasoning shift by shift, then confirm, dereplicate or flag an unknown.
A shortlist is useful only if it can be inspected. The live homepage preview shows the kind of objects a chemist reviews — not a guaranteed identity.
Ordered hypotheses with a match value so attention starts on the structures the data can actually support.
13C RMSD and evidence per shift make disagreements visible instead of hiding them in a single score.
The chemist confirms a known, dereplicates a mixture component, or flags an unknown. Spectra does not close the case.
01
Check whether submitted 1H and 13C data support a proposed structure.
02
Separate known library members from residual unknowns in extracts and mixtures.
03
Produce a defensible shortlist when the identity is not already assumed.
04
Rank plausible explanations for unexpected signals beside a known major component.
05
Support extract dereplication and analogue ranking before isolation campaigns expand.
06
Rank metabolite hypotheses from NMR with explicit residual uncertainty.
Spectra is intended to sit after acquisition and processing, not instead of them. The live product names Bruker, JEOL, Varian and JCAMP-DX as spectrum sources. That is workflow compatibility, not a certified connector to TopSpin, Mnova, Delta or ACD/Labs.
How Spectra fits with existing NMR software · Supported data
Spectra is the commercial product. ROSE is a public 1H NMR foundation model. Do not treat ROSE paper metrics as Spectra product accuracy.
Public repository for the 1H NMR foundation model.
Released ROSE-Pretrain-L checkpoint and model card.
ChemRxiv preprint for ROSE.
Zenodo archival record for ROSE releases.
Citation, model cards and the product versus research distinction
Spectra guides a laboratory from NMR data to an explainable shortlist of molecular identities. It retrieves plausible candidates, compares predicted and measured shifts, and returns match information, residual error, and evidence for chemist review. Ranking is not confirmation.
No. Spectra is a reasoning, ranking and evidence layer. Acquisition, processing, peak picking and reporting can remain in the tools already used in the laboratory.
No. The public product description includes verification of knowns, dereplication of mixtures, and identification of unknowns. Mixtures, impurities and overlapping signals remain harder cases and may require additional experiments.
No. The distinctive value is combining large-scale compound libraries with foundation-model ranking and physics-guided validation. Customer-private libraries may be discussed for enterprise pilots; that capability is not described as a default self-service feature here.
Both appear in the live product description. Verification of a proposed identity is usually the lower-risk path. Open-world identification of an unknown is possible but carries more residual uncertainty and more often needs a next experiment.
The current public product is an NMR compound-identification agent. Primary inputs stated on the live site are 1H and 13C data. Molecular formula, exact mass, MS or synthesis context may be supplied as additional context when available. Spectra is not positioned here as a standalone MS or IR product.
Candidate generation is not treated as sufficient. Hypotheses are ranked against retrieved structures and then checked by comparing predicted and measured shifts. Candidates the data cannot support should fall in rank or be discarded.
The live product surfaces a match value, 13C RMSD, and shift-level evidence. Confidence is conditional on the submitted data, the search space, and the quality of the spectrum. A high match is still a hypothesis until a chemist reviews it.
Spectra can support a documented review trail, but this website does not claim a completed validation package, 21 CFR Part 11 certification, or any other regulatory approval. Regulated use requires a separate enterprise discussion.
Self-service users can sign in and submit a spectrum from the live product. The fastest evaluations usually start with verification or dereplication on already-available 1H and 13C datasets. Enterprise pilots are scoped around a real case.
The chemist. Spectra produces a ranked shortlist and evidence. Final interpretation, reporting, and any quality or regulatory decision remain with the human expert.
Free includes 3 analyses. Pro is €79 per month or €790 per year. Lab is €299 per month for 5 users. Enterprise is custom. See the pricing page for who each plan is for. A paid plan does not confirm a molecular identity.