|Ph.D Student||Vitkin Edward|
|Subject||Computational Aspects of Metabolic Processes:|
Modeling, Analysis and Applications
|Department||Department of Computer Science||Supervisor||Professor Zohar Yakhini|
This PhD dissertation addresses various aspects of simulating the metabolic behavior of living organisms. The main topics include modeling and optimizing bio-industrial fermentation processes, the refinement of metabolic network models leveraging high-throughput knockout data and the normalization and visualization of data from clinical metabolite measurements.
We start with presenting BioLego, which is a freely downloadable webservice, ready for installation in the Microsoft Azure Cloud environment. BioLego provides a friendly and intuitive interface that enables the simulation and the optimization of single and two-step fermentation processes. The BioLego fermentation simulator is a scalable distributed framework, providing means to the process designers for analyzing, predicting and comparing the expected efficiency of several fermentation scenarios. It is based on a novel flexible modular modelling approach, enabling smooth generation of different multi-organism fermentation configurations consisting of independent encapsulated modules, representing individual organisms.
We continue with genome-scale single organism cellular metabolic models. Gene-fitness assays, measuring organism behavior in different growth conditions after genetic modifications, lead to newly generated data that are relevant for metabolic modelling. We developed a methodology to leverage these data to improve gene-to-reaction associations in existing genome-scale models. Specifically, we use these data to assign genes to orphan reactions. We also show how to integrate these data with statistics gathered from other types of functional-genomic data, such as gene-expression profiles, to facilitate gene-to-reaction assignments.
Finally, we present an approach for measurement-based analysis of human metabolism using the example of the steroidogenesis pathway. We describe a novel method for normalizing highly biased data, such as metabolomic hormonal profiles of children. We also describe a novel method for visualizing the analysis results both for patient-based and population-based perspectives.
The major contributions of this PhD research are as follows:
• BioLego project, including
o Novel modular modelling approach, enabling smooth generation of different multi-organism fermentation configurations
o Distributed framework based on Microsoft Azure Cloud environment, enabling high-scale simulations and optimization of single and two-step fermentation processes, including multiple knockouts in each organism.
o Intuitive interface, requiring no programming skills from potential users.
o Source code and installation instructions, freely downloadable and ready to install in the Microsoft Azure Cloud environment.
o Biological predictions for various fermentation scenarios supporting several published studies.
• Gene-to-reaction assignment methodology for fitness assays, including:
o A novel methodology to leverage fitness assays for genes and promoters to improve existing gene-to-reaction assignment techniques
o An integration of fitness assays statistics with other statistics based on gene-expression profiles
o Biological predictions of candidate genes potentially encoding proteins that catalyze 107 orphan reactions in iJO1366 metabolic model of E. coli.
• Analysis of clinical metabolomics profiles, including:
o A systems approach for analyzing steroid metabolomics data to identify a steroid-related disease signatures and patterns
o A novel peer-group normalization approach, clearly overperforming current methods in the field
o A novel visualization approach, addressing both patient-based and population-based studies
o Analysis of data from Congenital Adrenal Hyperplasia due to 21-hydroxylase deficiency and from childhood obesity subjects.