Improving constraint-based metabolic models with deep learning

Brasch, Brendan

Permalink

https://hdl.handle.net/2142/120576 Copy

Description

Title

Improving constraint-based metabolic models with deep learning

Author(s)

Brasch, Brendan

Issue Date

2023-05-02

Director of Research (if dissertation) or Advisor (if thesis)

Jensen, Paul A

Department of Study

Bioengineering

Discipline

Bioengineering

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

M.S.

Degree Level

Thesis

Keyword(s)

• deep learning
• metabolic modeling
• COBRA modeling
• constraint-based metabolic modeling
• genome-scale models
• neural networks
• Streptococcus mutans
• CobraNet

Abstract

Constraint-based metabolic modeling is a powerful tool that allows researchers to map out biological systems and run simulations in order to generate hypotheses related to metabolism. However, these systems are limited by their computational complexity and assumptions made during model creation that hinder predictive accuracy. Neural networks represent a promising approach that can be leveraged in combination with constraint-based models to elucidate further predictions from experimental data. However, researchers often have difficulty extracting features from neural network predictions, leading to difficulties generating and supporting research hypotheses. Furthermore, neural networks are limited by the availability and quality of relevant experimental data. Here, we describe a novel framework, CobraNet, which exploits the strengths of constraint-based models and neural networks in order to generate predictions. We demonstrate how this framework can be leveraged to generate predictions of enzyme activity and biological fitness on an understudied bacterium, Streptococcus mutans. This CobraNet model was built using a constraint-based model for S. mutans and simple metabolic experimental data, demonstrating the broad applicability of this framework. The CobraNet framework can be used throughout the field of biological modeling and would greatly enhance the accessibility of genome-scale models within the bioinformatics community.

Graduation Semester

2023-05

Type of Resource

Thesis

Handle URL

https://hdl.handle.net/2142/120576

Copyright and License Information

Copyright 2023 Brendan Brasch

Owning Collections