Drug Discovery AI
M

Matterworks

Matterworks, a Somerville, Massachusetts AI lab, sells Pyxis, a platform that reads high resolution mass spectrometry data, returns the identity and concentration of small molecules and lipids, and helps researchers work out what the changes mean biologically. Its Large Spectral Model, trained on more than 10 billion spectra spanning several million molecular structures, identifies compounds without a spectral library, and specialist models handle quantitation and interpretation. Through the Pyxis interface a scientist asks follow up questions, inspects the evidence behind an identification and refines the interpretation, for work such as target selection, drug response and biomarker hypotheses.

PyxisLabs, launched in 2026, sells it two ways. Labs running their own LC or GC mass spectrometry license the software from $25 per sample, excluding instrument operation, with standardized methods, SOPs, columns and universal calibrants included. The lab service at $200 per sample runs HILIC, reversed phase and lipid methods in positive and negative ionization and delivers identification, concentration and interpretation, with introductory projects from about $10,000 for up to 50 wells. Customer data and intellectual property stay isolated, and nothing a customer runs is folded into the shared foundation model.

In a July 2026 benchmark against the published AdipoAtlas reference for human adipose tissue, Pyxis recovered 394 of 505 lipids at its high confidence cutoff.

AI Health Index verifiedNovember 20, 2026
Compare Matterworks with other vendors
Founded
—
Headquarters
Somerville, Massachusetts, US
Categories
drug-discovery
Indexed Products
Pyxis, PyxisLabs de novo Chemotyping
Buyer Segments
Biopharma, Biotechnology Company, Academic Research Lab, Translational Research Team
Assessment

Capability Axes

An AI Health Index grade measures what a buyer can verify from public sources on the date shown. It is not a rating of how good the product is. A vendor can build an excellent system and grade low on an axis because it publishes nothing an outsider can check. How grades read

AI Capability
AA on AI CentralityThe artificial intelligence is the product. Remove the model and there is nothing left to sell.
Vendor Published

Pyxis runs on Matterworks models that turn instrument measurements into estimates of molecular identity and concentration, identifying compounds without a spectral library, and researchers then use the interface to explore how measured changes might relate to a biological explanation. A large spectral model handles the instrument data, with specialist models for particular tasks.

The lab service collects measurements when customers send samples and licensed customers bring their own data, but on either route the models produce every identification, concentration and interpretation the customer pays for.

BB on Autonomy and Oversight ModelThe oversight structure is described and one part is missing, commonly the threshold at which the system stops or what happens after it is wrong.
Vendor Published

A scientist supplies the data and the question, reviews candidate identifications and interpretations, and asks follow up questions to decide how much confidence a result deserves. The AdipoAtlas work reports confidence categories, so findings can be inspected at different cutoffs, and Matterworks acknowledges that predictions can be wrong. The research product directs no patient care and writes nothing to a health record. No escalation threshold, automatic hold or procedure for reversing an exported interpretation is described.

BB on Model and Technology TransparencyThe approach or the suppliers are named without the version and update discipline behind them. Naming a supplier is the entry to this band both here and on Model Supply Chain Disclosure, which ask different questions of the same disclosure: who receives the data, and what produces the output.
Vendor Published

Matterworks names its Large Spectral Model and pairs it with models trained for specific interpretation tasks, and it characterizes the foundation model's training data as more than 10 billion spectra spanning several million molecular structures and multiple biological contexts. The measured input is kept apart from the interface a scientist questions, and earlier work sets out the approach to quantitative predictions from instrument data. A buyer cannot tie a release to a frozen training set, read version histories for each component model or rely on a release schedule or change notice for enterprise customers.

BB on Model Supply Chain DisclosureSubstantial partial disclosure, or a chain that is structurally short: an in house build, a cleared model that cannot be quietly swapped, or a deployment where the transfer does not occur at all. Naming the model provider exits the band below into this one; the axis rises from here on the completeness of the party list and on the terms that govern data once it arrives.
Vendor Published

Matterworks builds the Large Spectral Model itself as the foundation of Pyxis, and enterprise programs can run models dedicated to one customer. Pyxis also uses an interactive language interface and further expert models, but the provider of that language interface, any inference vendor and the storage provider are not named. No list shows which parties can receive a spectral file or a scientist's question, or what retention terms apply to them, so the in house foundation model covers only part of the chain.

BB on Clinical and Operational EvidenceNamed deployments with dated outcome figures and enough method to test them, or published research short of independent validation.
Vendor Published

In a July 2026 application note, Matterworks benchmarked Pyxis on part of AdipoAtlas, a published reference for human adipose tissue built by an outside research team. Pyxis recovered 394 of 505 reference lipids at its high confidence cutoff and 431 with the lower confidence categories included. Matterworks ran the comparison itself, and a 2024 preprint with academic collaborators tested earlier Pyxis quantitation against conventional reference measurements. No independent replication of the full commercial workflow and no prospective study of the current product exist yet.

BB on AI Safety and PHI StewardshipCategorical commitments are published, such as no training on customer data, without the retention schedule behind them.
Vendor Published

Commercial licenses keep customer data and intellectual property isolated, and nothing a customer runs, spectra or sample information, is folded into the shared foundation model, including when PyxisLabs runs the analysis as a service. Enterprise models tuned to one organization's biology serve no other customer. No deletion schedule covers datasets or backups, there is no de identification rule for inputs from human research, and no procedure covers an unsafe interpretation.

Regulatory and Compliance
DD on HIPAA and BAA PostureNo statement of status and no privacy document that reaches the product. Both halves are absent at this grade: a service level privacy document on its own lifts a vendor to the band above even where no status is stated. A website privacy notice, including one that says it does not cover customer data, does not reach the product, so a vendor whose only document is that notice and who states no status grades here.
Vendor Published

The Matterworks privacy policy covers visitors and contact information collected through its sites, and a Pyxis enterprise account runs under a separately negotiated service agreement. Clinical research teams can bring instrument data from human studies, yet no business associate agreement or status is stated and nothing defines when Pyxis may accept protected health information.

The isolation and no training commitments answer a confidentiality question rather than this one, and the handling of identifiable human sample records and the terms binding service providers are not stated.

DD on Security Certifications and Trust CenterControls are asserted with nothing independent behind them, or nothing is published. Read the note before concluding anything: this is the grade most often corrected on a second pass, because assurance material frequently sits on a parent domain or inside an old announcement rather than on the product pages.
Vendor Published

Matterworks promises enterprise security and compliance for Pyxis and makes data isolation part of every engagement, with enterprise tuned models serving no other customer. Its site privacy policy covers safeguards for website personal information and concedes that no electronic storage or transmission is fully secure.

No audit report, attestation against a named framework or independent test scope covers the Pyxis service, so its audit period, certification boundaries and provider controls cannot be confirmed from outside.

BB on FDA and Regulatory StatusThe pathway is stated and in progress, or a clearance is named without the vintage and scope a buyer needs to match it to the product on offer. A product outside device regulation also sits here when the vendor publishes what it is for and in what setting and the product on offer matches, or names another regime it works under, such as a registration outside the United States. A general clause saying the service is not medical advice does not reach this band.
Vendor Published

The Pyxis license is offered for life science research, such as studying drug response, generating biomarker hypotheses and testing how measurements bear on research objectives. Published evaluations measure analytical identification or quantitation against reference data, sometimes in human samples.

Neither offering presents a test for diagnosing patients or directing treatment, and there is no clinical diagnostic label or device clearance, which matches a research product sold to supply measurements and interpretations to research teams.

DD on AI Governance and Bias DisclosureNothing published on how model behavior is governed or tested. Multilingual operation with no subgroup performance sits here when the vendor markets recognition quality as a strength, because a caller the system failed to understand leaves no complaint and no record.
Vendor Published

The AdipoAtlas benchmark separates outputs by confidence and measures recovery against a curated reference, with recovery varying across fractions of that reference set, a test of which lipids the model finds in one tissue rather than of how it behaves across populations or instruments. Dedicated enterprise models keep one customer's work confidential. There is no review process for fairness, no approval plan for changes to model behavior and no study of error across the human populations, instruments and materials a new deployment might bring.

CC on AI Liability and RecourseMechanisms exist that let someone challenge an output, such as audit trails, source traceability or review before commit, with nothing standing behind the output and no route for the harmed party.
Vendor Published

Scientists can challenge a proposed identification by inspecting its supporting evidence and asking Pyxis to reconsider, and commercial arrangements include priority support and solution engineering. Matterworks says plainly that Pyxis, like a human scientist, can make mistakes. The website terms cover the site and leave Pyxis to a separate customer agreement, and no accuracy guarantee, indemnity or duty to correct a scientific result is offered outside it.

Integration and Deployment
DD on EHR and Interoperability DepthNo integration evidence. A connector described as available on request grades here until one exists.
Vendor Published

Pyxis takes LC and GC mass spectrometry data, from a lab's own high resolution instruments or the PyxisLabs service, and returns measured and interpreted results for research. No FHIR or HL7 interface, EHR marketplace listing or exchange of clinical records with a hospital exists for it. A sample may come from a study that enrolled patients, but the product handles research data and its context, never individual clinical records.

CC on Deployment Model and Data ResidencyA single hosted option with location implied rather than committed.
Vendor Published

Pyxis was built as a cloud platform, and the current offer separates each customer's data and intellectual property, with enterprise models that serve only the organization that commissioned them. A researcher can run the instrument on site and use Pyxis to interpret the data, which places the measurement but not the model or storage. The hosting region, backup location, residency terms for customer files and any on premises installation are not stated, and the isolation commitment says nothing about where inference runs.

Commercial
BB on Commercial TransparencyA price or a pricing basis is published without full tiers, so a buyer can size the cost before making contact.
Vendor Published

Pyxis licenses start at $25 per sample for analyzing data from a customer's own compatible instrument, excluding instrument operation, and PyxisLabs charges $200 per sample to measure and interpret a physical sample, with introductory and demonstration projects from about $10,000 for up to 50 wells. Both start through sales contact, an enterprise agreement can change what is included, and annual license commitments, special integrations and volume discounts carry no standard price.

BB on Setting and Specialty CoverageCoverage is named with validation behind part of it.
Vendor Published

Analytical labs and life science researchers working with small molecules and lipids use it, including drug discovery and translational teams, either on their own compatible instruments or by sending samples to PyxisLabs. The AdipoAtlas benchmark on human adipose tissue gives one concrete setting where the approach was tested, and an earlier study examined absolute quantitation in other research materials. The research coverage stops short of clinical diagnosis, and the tests so far do not reach every sample type or disease the platform could be used to study.

Alternatives

Alternatives to Matterworks

Other products that do the same kind of work as Matterworks, and what each one does.

Bexorg tests drugs for central nervous system diseases in donated postmortem human brains before they reach clinical trials, working for pharmaceutical and biotechnology sponsors and on its own licensed candidates.

Numerion Labs, formerly known as Atomwise, is a San Francisco AI drug discovery company developing small molecules for immune and inflammatory disease, led by chief executive Steve Worland.

Enveda is a clinical stage biotechnology company in Boulder, Colorado that uses AI to discover medicines from the chemistry of living things.

b12 Labs sells AI software to pharmaceutical and biotechnology chemists planning small molecule synthesis.

Cellarity is a Flagship Pioneering company in Somerville, Massachusetts, founded in 2019, that designs drugs against the behavior of a whole cell rather than against a molecular target.

Variational AI develops Enki, a generative model for designing small molecules against a research team's target product profile.

Commercial

Pricing

Vendor-published figures are labeled as such. Figures labeled “Estimated” are derived from third-party sources and have not been confirmed by the vendor.

Entry Price Pricing Basis BAA Tier Implementation Source
$25 per sample
$25 baseline
Subscription and enterprise licensing starts at $25 per sample using customer instrument data, excluding instrument operation. The separate laboratory service is $200 per sample. No business associate agreement or protected health information tier is stated. The lab service FAQ gives a floor of approximately $10,000 for introductory or demonstration projects covering up to 50 wells. License minimums and integration fees are not listed. Vendor Published

Licensing and lab service orders start through sales contact. The prices cover identification and quantitation for small molecules and lipids.