Skip to content

Absolute Quantitation of Critical Bioprocessing Metabolites with the AI-​powered Pyxis Metabolomics Platform

Absolute Quantitation of Critical Bioprocessing Metabolites with the AI-​powered Pyxis Metabolomics Platform

Biopharma is seeking new strategies and innovations to improve process efficiency across biomanufacturing, with the aim of producing therapeutic proteins in a cost-effective and consistent manner. Metabolomics characterizes the cell biochemistry that determines the efficiency of a bioprocess. As such, metabolomics has shown great promise as a critical tool and readout for optimizing bioprocesses, in some cases enabling the increase in cell specific productivity (qP) by 490%1. Yet, adoption of metabolomics is hampered by the difficulties in measuring both intra- and extracellular metabolites quantitatively and in-loop in cell line development, clone selection, and process development2.  

Matterworks, through the use of deep learning, simplifies the way absolute metabolite concentrations are determined. The Matterworks PyxisTM metabolomics platform quantitates metabolite concentrations from raw mass spectrometry (MS) data, without the need for method development or calibration curves. Pyxis integrates into existing LC-MS equipment and sample preparation to absolute concentrations takes less than 10 minutes per sample. 

Pyxis provides:

  • Absolute concentrations of 46* intracellular and extracellular metabolites 
  • Validated LC-MS methods 
  • A Scalable solution enabling low to high throughput screening

    * menu will grow to 150+ in 2023

Pyxis-LC-MS Methods:

The platform utilizes readily available LC-MS technology. Here, we used a Thermo Fisher Transcend LX-2/LX-4 connected to a Vanquish Flex, enabling parallel column equilibration and injection. Data was recorded on two Thermo Orbitrap Exploris 120 mass spectrometers. 

Sample prep to results in 10 minutes 

ML/AI enables MS to metabolite concentrations in minutes:

The process by which an analyte is ionized prior to introduction into a mass spectrometer involves a myriad of complex interacting parameters that are known to affect ionization efficiency3. Because of this complexity, efforts to build physics-based models to predict ionization efficiency have largely failed. In the last several years, significant advances in machine and deep learning have emerged that are well-suited to model high-dimensional, nonlinear phenomena such as MS ionization4.

The data below illustrates the natural advantages of a deep learning technology platform, and demonstrates the opportunity for metabolomics to become a routine measurement in cell line development, clone selection, and process development.

Experimental Design and Results:

The Pyxis Alpha prototype was tested to demonstrate that a deep learning model could be used to derive quantitative metabolite concentrations from three bioreactors run at Keck Graduate Institute (KGI). Each bioreactor represented different feeding strategies focused on maintaining constant glucose (black line) or glutamine (red line) concentrations over the culturing period. The third bioreactor (blue line) served as the negative control representing a typical batch-fed reactor with a single feed at the beginning of the culturing period. Raw MS data was fed into Pyxis and metabolite concentrations were generated in a few minutes, without the need to baseline pick peaks or run calibration curves. To validate the Pyxis results, traditional LC-MS calibration curves were generated and the data was correlated to the results from Pyxis.

Figure 1: Extracellular Metabolite Concentrations generated from Pyxis across three Bioreactors

Figure 2: Intracellular Metabolite Concentrations generated from Pyxis across three Bioreactors 

Figure 3: Correlation of concentrations generated from Pyxis to MS-matched isotopologues

Figure 4: Correlation of concentrations generated from Pyxis to Cedex Bio HT instrument

Summary of Alpha Pyxis results: 

  • Concentration accuracy that meets or exceeds the typical performance obtained using matched isotopologues, per-run calibration, and expert data interpretation
  • High correlation of absolute quantitation between Pyxis and commonplace at-line benchtop instruments (e.g. Novaflex, Cedex Bio)
  • Quantitative precision, demonstrating excellent reproducibility with an observed median RSD of 5%
  • Rapid quantitation of intra- and extracellular metabolites without complex method development and standard curves 


This Alpha prototype data validates the use of machine learning for the quantitative analysis of the intracellular and extracellular metabolites that are critical to the bioprocessing industry. Literature has shown that metabolomics can provide insight into intracellular pathways predictive of cell growth, protein production, and cellular health, which enables everything from faster clone selection to significant gains in cell-specific productivity and titer. To date, however, current metabolomics solutions are either limited in scope or require extensive expertise in method development and post-machine analysis. Comprehensive metabolomics are therefore traditionally used as a post hoc analysis, rather than a standard tool for process development. By integrating an ML-based workflow into traditional LC-MS analytical technologies, the Pyxis metabolomics platform offers a comprehensive, real-time metabolomics solution for the interrogation of cellular health, energy production, and redox potential. Absolute quantitation enables cross-metabolite and cross-sample comparisons, regardless of MS instrumentation, site, or experiment, and greatly improves process scale-up, process consistency, and successful site transfers.  

Request Seminar

Acknowledgments: Matterworks acknowledges Christine Urrea and Dr. Hu Zhang from the Keck Graduate Institute for providing technical support and samples for analysis.

  1. Yao G, Aron K, Borys M, Li Z, Pendse G, Lee K. A Metabolomics Approach to Increasing Chinese Hamster Ovary (CHO) Cell Productivity. Metabolites. 2021 Nov 30;11(12):823. doi: 10.3390/metabo11120823. PMID: 34940581; PMCID: PMC8704136.
  2. lseekh, S., Aharoni, A., Brotman, Y. et al. Mass spectrometry-based metabolomics: a guide for annotation, quantification and best reporting practices. Nat Methods 18, 747–756 (2021).
  3. Liigand, Jaanus, Anneli Kruve, Piia Liigand, Asko Laaniste, Marion Girod, Rodolphe Antoine, and Ivo Leito. A2015. “Transferability of the Electrospray Ionization Efficiency Scale between Different Instruments.” Journal of the American Society for Mass Spectrometry 26 (11): 1923–30.
  4. Mizuno, Eiji Sugiyama, Toshimasa Toyo’oka, and Kenichiro Todoroki. 2021. “Machine Learning Guided Prediction of Liquid Chromatography-Mass Spectrometry Ionization Efficiency for Genotoxic Impurities in Pharmaceutical



Get the Latest

Subscribe to our blog