Computational strategies for metabolic modeling of yeast

Metabolic engineering through genetic modification offers enormous potential for redesigning metabolic pathways of microorganisms, with the objective of overproduction of a desired product. However, a more in-depth, comprehensive, and systematic understanding of metabolism and the impact of genetic modification on metabolism is required. Computational modeling offers one route for systematic investigation of the complex pathways within a microorganism and how they respond to perturbations. This dissertation focuses on developing new modeling methods to build a deeper understanding of yeast, an industrially relevant unicellular microorganism. Mechanistic kinetic models and deep-learning models were constructed to study the lipid metabolism of the model yeast S. cerevisiae and the non-model oleaginous yeast R. toruloides. The kinetic model for S. cerevisiae was trained on lipidomic data generated using a high-resolution Orbitrap mass spectrometer. The kinetic model provided valuable insights, such as the presence of a futile cycle in S. cerevisiae, which can be accounted for in future engineering strategies to avoid carbon being trapped in such cycles. Next, the study of lipid accumulation in the oleaginous yeast R. toruloides was studied in a multi-omic integration approach, in collaboration with Dr. Anshu Deewan from Prof. Christopher Rao’s lab. Two different ‘omics datasets, namely transcriptomic and lipidomic analyses, of lipid accumulation were studied in an integrative manner to identify the regulatory origins of lipid accumulation in R. toruloides. Then, deep learning models and techniques were developed in collaboration with Michael Volk for assisting in two new applications of metabolic modeling: (i) graph neural networks to capture the network topology and concentration measurements of in silico metabolic networks, and (ii) training a conditional variational autoencoder (CVAE) on parameter estimation data generated from the development of kinetic model of S. cerevisiae. The deep-learning model showed good recapitulation of synthetic data in some cases, specifically from the application of graph attention networks to model metabolic fluxes. The CVAE accurately captured the parameter landscape of the kinetic model and was thus capable of suggesting new parameter vectors to train the kinetic model. Finally, the modeling paradigm for the S. cerevisiae kinetic model was adapted for R. toruloides, which can natively accumulate much higher titers of lipids than S. cerevisiae but has not been extensively reported in literature. The computational methods developed herein provide new tools for understanding complex metabolic pathways to guide industrial applications of metabolic engineering.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Chemical Engineering

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2023

Author dc:creator

• Mishra, Shekhar

Contributors dc:contributor

• Zhao, Huimin
• Rao, Christopher V
• Shukla, Diwakar
• Sinha, Saurabh

• Mathematical Modeling
• Yeast
• Metabolic Engineering

Statement dc:rights

• Copyright 2023 Shekhar Mishra

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/120087

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F120087