Identify your MS/MS spectra with classical and ML scoring in one workspace.
MSMS Analyzer brings cosine, modified-cosine and neural embedding models (Spec2Vec, MS-BERT, DreaMS) together over your reference libraries — with interactive mirror plots that make every match auditable.
No sign-up wall, no payment. We onboard a small number of labs by hand.
Built on open, peer-reviewed foundations
MGF · MSP · mzML · JSON
classical + ML
GNPS · MassBank · MoNA · MS-DIAL · LIPID MAPS
The problem
Spectral identification is scattered across tools that don't talk to each other.
Metabolomics teams juggle one script for cosine scoring, another for embedding models, a viewer for mirror plots, and yet another importer per library. Comparing methods on the same spectrum — and trusting the answer — takes far too long.
- Classical scores and ML models live in separate, hard-to-reproduce pipelines.
- Mirror plots are generated ad-hoc, so matches are hard to audit and share.
- Every reference library ships its own format and import quirks.
What it does today
One workspace, from raw spectrum to auditable match.
The metabolomics core is implemented and tested. We describe below what works now — not a wish list.
Multi-format ingestion
Upload MGF, MSP, mzML or JSON. Peaks are parsed, normalized and ready to score in seconds.
Classical + ML scoring side by side
CosineGreedy, ModifiedCosine and neutral-loss scores next to Spec2Vec, MS-BERT and DreaMS embeddings — on the same spectrum.
Interactive mirror plots
Every candidate match renders as a mirror plot you can zoom, inspect peak-by-peak, and share — so identifications are auditable, not black-box.
Reference-library search
Import and query GNPS, MassBank, MoNA, MS-DIAL and LIPID MAPS, grouped by molecule (InChIKey) for fast triage.
Analog & fingerprint search
Vector similarity search over embeddings surfaces structural analogs even when an exact library hit is missing.
Interpretation copilot
An optional assistant helps read results in context — summarizing why a candidate scores the way it does.
How it works
Four steps from upload to identification.
- 01
Upload
Drop your MS/MS spectra in any supported format.
- 02
Choose methods
Pick the classical scores and ML models you want to compare.
- 03
Match
The engine scores against your libraries and groups hits by molecule.
- 04
Audit
Read each match as an interactive mirror plot and export what you keep.
Who it's for
Built for the people doing the identification.
Metabolomics labs
Small-molecule identification against spectral libraries — the mature core of the platform.
Proteomics teams
DIA peptide/protein workflows — in progress, and a natural fit for early partners with DIA data.
On the roadmap — shaped with partners
Where partner labs steer us next.
These capabilities are in progress. Design partners decide their priority and requirements.
DIA proteomics workflows
Guided DIA analysis (alphaDIA today; DIA-BERT for partners running their own GPU + licensed weights).
Batch & reporting
Higher-volume batch runs and exportable, reproducible reports for regulated environments.
Your requirement here
The point of the program: the first partners' needs set the near-term roadmap.
The partner program
We're building this with a handful of labs — not selling it yet.
MSMS Analyzer is in active development. Instead of a public launch, we're onboarding a small number of design partners who get early access and, in return, shape what we build next.
No payment is taken during the partner program. This page is not a checkout — it's an introduction.
Early hands-on access
Use the platform on your own spectra, guided and supported directly by the team.
Roadmap influence
Your workflows and must-haves set our near-term priorities.
Co-development
We build alongside you, iterating on real data and real constraints.
Founding-partner terms
Preferential launch terms when the product is ready — decided together, never charged upfront.
Partner application
Tell us about your lab.
A short form. We read every submission and reply personally.
Thank you — application received.
We'll be in touch personally at the email you gave us. In the meantime, feel free to reply with any data specifics.
FAQ
Straight answers.
Is the product finished?
No — and we won't pretend otherwise. The metabolomics identification core works and is tested; parts of the commercial and proteomics tooling are still in development. The partner program exists precisely to build the rest with real labs.
What does being a design partner involve?
Early, hands-on access to the platform, a few conversations about your workflows, and feedback as you use it on your own spectra. You help set priorities; we do the building.
Does it cost anything?
No. We take no payment during the partner program. If and when the product is ready to sell, founding partners get preferential terms decided together — never billed upfront here.
What happens to my data?
During the program we only use the information you submit to evaluate and follow up on your application. Any analysis data you share as a partner is handled under terms we agree with you directly.
Which methods and libraries are supported?
Today: classical scores (CosineGreedy, ModifiedCosine, neutral-loss) and ML models (Spec2Vec, MS-BERT, DreaMS), over GNPS, MassBank, MoNA, MS-DIAL and LIPID MAPS. DIA-BERT proteomics needs a licensed model and GPU and is offered only to partners set up for it.