PyxisLabs removes the per-analyte calibration bottleneck from untargeted LC-MS metabolomics with full-service and enterprise licensing routes at sequencing-scale economics
Quantitative untargeted metabolomics has long demanded something it could rarely sustain: the broad chemical coverage of an untargeted experiment combined with the concentration precision of a standard curve built for every analyte in the run. The dependency on physical calibration standards for each target molecule has kept per-sample costs high and throughput low, constraining how far untargeted LC-MS can scale across large population cohorts or high-throughput screens. Matterworks, Inc. responds to that constraint with two commercial offerings under PyxisLabs: a full-service biochemical analysis route priced at $200 per sample, and enterprise licensing of the Pyxis co-scientist for approximately $25 per sample on a laboratory's own LC-MS instruments.
What is the calibration bottleneck in untargeted metabolomics?
The calibration bottleneck is the requirement to build a separate standard curve for every analyte a laboratory wants to quantify. In practice, this makes absolute concentration measurement in untargeted experiments impractical at scale, forcing most workflows to report relative abundances instead. That substitution limits comparability across sites, time points, and instrument platforms in ways that annotation software cannot fix, because the problem is upstream of annotation: it lies in the measurement step itself.
Working in analytical science?
Register for a FREE Separation Science account to subscribe to the Separation Science Newsletter.
Quantitative interpretation remains the persistent bottleneck even after software handles feature detection and peak alignment. Matterworks addresses this with its Large Spectral Model (LSM), a foundation model that interprets small-molecule, lipid, and peptide mass spectrometry data directly from raw instrument readouts. The LSM predicts molar concentrations computationally, replacing per-analyte empirical calibration with algorithmic inference of ionization efficiency and matrix effects.
The approach connects to the quantitative rigor the FDA's ICH M10 Bioanalytical Method Validation and Study Sample Analysis guidance establishes for consistent, reproducible concentration measurement across complex biological matrices. Meeting those expectations without per-compound calibration overhead requires methods that generalise across analyte classes, which is the same problem large spectrum models now tackle directly on LC-MS and GC-MS spectral signals through self-supervised learning rather than curated reference libraries.
Removing the calibration dependency computationally is what Matterworks argues allows untargeted metabolomics to scale the way sequencing did: by decoupling coverage from the cost of per-analyte physical standards.
How does PyxisLabs deliver untargeted metabolomics at $25 per sample?
PyxisLabs offers two access routes: a full-service laboratory that accepts sample submissions for de novo chemotyping at $200 per sample all-in, and an enterprise Model-as-a-Service (MaaS) subscription that runs the Pyxis co-scientist on a laboratory's own LC-MS instruments at approximately $25 per sample. Facilities that submit samples to the PyxisLabs AI Native Laboratory receive small-molecule and lipid identification, untargeted concentration determination, and biological interpretation from the Pyxis co-scientist. Research organizations that license Pyxis for their own instruments run interpretation locally, keeping raw and structural data on-site.
| Biochemical analysis services | Full-service sample submission to PyxisLabs |
| Enterprise licensing | Subscription software deployed on customer instruments |
| Services pricing | $200 per sample |
| Licensing pricing | ~$25 per sample |
| Services outputs | De novo chemotyping, molar concentration determination, biological interpretation |
| Licensing outputs | Raw data interpretation, untargeted quantitation, scalable module deployment |
Application areas the company identifies include:
- Exposomics characterization and biomarker discovery
- Quantitative mechanism of action studies
- Target discovery, lead identification, and optimization
- Large population cohort studies
- High-throughput screening
Biopharmaceutical and contract research organizations running high-throughput screening programs fit the enterprise licensing tier, where per-sample economics determine whether a study is feasible at cohort scale. Institutions without in-house mass spectrometry, or those needing to expand untargeted metabolomics capacity quickly across clinical or population-scale sample sets, suit the full-service route.
What Matterworks says about the PyxisLabs launch
"Sequencing scaled the moment that its cost and turnaround stopped being a constraint: biology is now foreseeable at similar cost and scale for the biochemical layer," notes Fadi Abdi, Ph.D., Executive Vice President and Head of PyxisLabs at Matterworks. "Launching biochemical analysis services and offering Model-as-a-Service licensing enables every lab to choose how it engages Pyxis: send us the sample or run Pyxis on the instrument you already own."
Amy Caudy, Ph.D., Vice President, Biochemical Omics at Matterworks, adds: "Pyxis puts PhD-level biochemical interpretation directly in the hands of the researchers who need it. A lab can license the modules it needs and immediately apply Pyxis to applications such as biomarker discovery, quantitative mechanism of action, target discovery, lead identification and optimization, or large population studies."
Small-molecule and lipid modules are available immediately, with a peptide module scheduled for Q4 2026. The two-tier commercial structure gives laboratories the option to engage PyxisLabs through sample submission or instrument-level licensing, depending on existing infrastructure. Both routes draw on the same LSM-based computational approach to untargeted metabolomics quantitation.
This article is based on a press release issued by Matterworks, Inc. and was produced under Separation Science's AI Editorial Guidelines.
