Sponsored content brought to you by
Rapid determination of the concentrations of endogenous metabolites in biological systems is of unquestionable scientific value across applications such as drug discovery and development, microbiome science, synthetic biology, health, nutrition, and agriculture. Yet, it remains a holy-grail problem in the analytical sciences.
Absolute quantitation offers objective measurements that connect directly to translational biology, kinetics, and phenotype in a way that relative quantitation simply does not. Furthermore, grounding measurements in absolute concentration units inherently provides seamless comparability across data sets acquired at different times, on different instruments and across different studies or experiments.
To address these challenges, we harnessed recent breakthroughs in machine learning to reimagine how LC-MS–based untargeted metabolomics is performed. Our approach enlists large-data semantic models and transformer-based architectures, to learn the quantitative relationship between raw MS data and the concentrations of the molecules present in the sample.
Pyxis is an AI-enabled technology for absolute quantitation of untargeted analytes that: (1) standardizes analysis to remove custom method development; (2) allows non-experts to rapidly determine absolute analyte concentration by eliminating highly specialized tasks; and (3) return concentrations for an increasingly broad set of analytes.
Quantitative Metabolomics—The Problem
LC-MS is a spectacularly powerful tool for the detection and identification of biologically critical metabolites. While peaks in an untargeted LC-MS chromatogram can be associated directly with molecular identity, integrated peak area is only indirectly associated with concentration.
The functional output differences between relative and absolute quantitation are demonstrated in Figure 1—which shows relative quantitation in panel A and absolute concentration in panel B. The x-axis in both panels is the true concentration of 50 analytes. In Figure 1A, changes in concentration are captured as increases in peak area, yet both the experimentally observed abundance (peak area), and the scale of the change (slope), are deeply analyte dependent.
In contrast, the y-axis in Figure 1B is in micromoles per liter (µM)–a universal and reproducible scale. Even more powerful than the experimental utility of global comparability, absolute metabolite concentration is a deeply biologically relevant quantity. Absolute concentration can be used to gain better understanding of enzyme kinetics to model pathways and in the development and validation of mathematical models of metabolism.
While the power in directly reporting absolute concentrations of metabolites is evident, in conventional workflows, converting peak area to concentration is achieved by explicit analyte-by-analyte calibration. Absolute concentration is thus limited to narrow panels of selected molecules for which a method is developed, and standards are available.
Introducing Pyxis
The first application of the Pyxis technology is focused on the identification and quantitation of polar metabolites. It includes five critical technology pillars—(1) universal calibrators (StandardCandles™) to represent chemical space of polar metabolites, (2) a turn-key LC-MS method optimized for speed and broad, sensitive detection, (3) an immense, well labeled internal training data set used as input for (4) the unique Pyxis ML model, and (5) the secure cloud-based software platform.
The model is constructed to predict absolute concentration based on generic chemical structure, not specific analyte behavior and is thus functionally an untargeted assay designed to identify and quantify a broad range of analytes, rather than a targeted assay where the results are applicable to a small subset of analytes.
Matterworks has recognized the immense power of absolute quantitation to metabolomics and has reimagined what is possible from raw LC-MS data. In a truly interdisciplinary effort, we are taking powerful AI techniques developed for language and image processing and applying them to LC-MS data, enabling the direct transformation of uninterpreted, raw LC-MS data to a biologically actionable list of the identities and concentrations of detected metabolites.
Learn more www.matterworks.ai/pyxis.
