Combining Chemometrics and non-target GC-IMS gas phase analysis for improved bioprocess monitoring

Industrial biotechnology uses microbial fermentation processes to develop or modify products and is a major contributor to the production of high-value goods, including pharmaceutical active ingredients, feed nutrients, and flavor and fragrance compounds. Process Analytical Technology (PAT) is essential to evaluate batch performance and ensure product quality. It aims to automatically measure as many parameters as necessary, in real time and to gain process insight through chemometric data analysis. Improving process understanding through novel and advanced analytical technologies is an active area of research. In bioprocess monitoring, most established PAT focuses on the liquid broth. A potential addition to established PAT is the analysis of volatile organic compounds (VOC) including microbial VOCs (mVOC), in the fermentation headspace. The hypothesis was, that mVOC undergo qualitative and quantitative changes during the dynamic fermentation process and can be correlated to parameters that are difficult to measure directly. Gas chromatography hyphenated to ion mobility spectrometry (GC-IMS) is an emerging analytical platform for trace-level VOC analysis. GC-IMS is particularly known for combining sensitivity, and selectivity due to two-dimensional separation with a robust and reliable setup, ideal for point-of-care use. The principal objective of this study was to examine the potential of GC-IMS-based VOC measurements in conjunction with chemometric non-target screening techniques as a soft sensor for bioprocess monitoring. In consequence of the lack of suitable software, a Python package for the multivariate analysis of GC IMS data was developed and subsequently published as open-source software under the name gc-ims-tools. It implements essential, data-specific functionalities, including file readers, visualizations, preprocessing, dataset organization, and workflows for common chemometric algorithms. A significant challenge in the analysis of GC-IMS data is the high dimensionality and collinearity. Two strategies were investigated in detail to address this question. The suitability of different dimensionality reduction and feature selection methods were evaluated as suitable preprocessing steps for a variety of machine learning algorithms and interpretation purposes. In the context of classification tasks, filtering based on PLS variable importance in projection scores proved to be an effective variable selection approach. The second approach, to reduce the dimensionality, was to extract peak lists from the raw data. Persistent homology was proposed as a suitable algorithm for automated peak detection in two-dimensional GC-IMS data. Its performance was evaluated on two publicly available datasets with different GC setups, peak shapes, and resolution. The utilization of certain preprocessing methods, mainly asymmetric least squares baseline correction was found to improve the number of correctly detected peaks. Two studies on exemplary processes demonstrated that GC-IMS-based VOC measurements do contain information about the process. In the first experiment, shake flask cultures of E. coli, S. cerevisiae, L. brevis, and P. fluorescens were prepared and monitored with an offline GC-IMS system following headspace incubation of a liquid sample. In addition to pure cultures, mixed cultures with combinations of two organisms were measured as simulated contaminations. The microorganisms were successfully classified by PLS-DA and the mixed cultures could be distinguished from the pure cultures. The results are a promising first step towards headspace contamination detection. Furthermore, the optical density, as a surrogate for biomass, could be predicted with the gas-phase measurements and support vector regression (SVR). An industrial, continuous Bacillus licheniformis process was monitored. The online data can be used for both, targeted tracing of specific compounds and combined with chemometrics as non-target “fingerprints” characteristic of more abstract variables, such as batch maturity. Some of the peaks detected showed oscillating concentration variations, not present in other PAT and not caused by the process control parameters, such as feed rate. In conclusion, GC-IMS-based VOC measurements proved to be a promising novel process analytical technology for bioprocess monitoring with high potential for future application development.

Author dc:creator

• Christmann, Joscha

Advisor dc:contributor.advisor

• Rohn, Sascha

Licence dc:rights.uri

• http://rightsstatements.org/vocab/InC/1.0/

Language dc:language.iso

en

Identifier URI

https://doi.org/10.14279/depositonce-23581

OAI identifier oai:identifier

oai:depositonce.tu-berlin.de:11303/24765