Application of machine learning for the creation of a gut microbial biomarker panel enabling personalized nutrition recommendations

Shinn, Leila Marie

Permalink

https://hdl.handle.net/2142/120548 Copy

Description

Title

Application of machine learning for the creation of a gut microbial biomarker panel enabling personalized nutrition recommendations

Author(s)

Shinn, Leila Marie

Issue Date

2023-04-24

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

Holscher, Hannah D

Doctoral Committee Chair(s)

Donovan, Sharon M

Committee Member(s)

• Madak-Erdogan, Zeynep
• Rodriguez-Zas, Sandra L
• Zhu, Ruoqing

Department of Study

Nutritional Sciences

Discipline

Nutritional Sciences

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Gastrointestinal Microbiota
• Metabolomics
• Metagenomics
• Multiomics
• Fidelity Measures
• Dietary Intake Biomarker
• Machine Learning
• Multiclass

Language

eng

Abstract

Diet affects the human fecal microbiota, metagenome, and metabolome. This effort aims to identify a compact set of highly predictive objective microbial biomarkers of food intake utilizing pre- and post-intervention fecal samples from five completed clinical trials examining almond, avocado, broccoli, walnut, and whole-grain barley and oat intake. First, we used random forest models to predict consumption of these six foods with 70% accuracy using 22 taxa from 16S rRNA sequencing data. Collapsing barley and oats into 1 grains category increased the model accuracy to 77% (23 taxa). Overall model accuracy was 85% (15 taxa) for classifying almond (76% accurate), avocado (88% accurate), walnut (72% accurate), and whole grains (96% accurate) intake (1)(Chapter 3). Second, metagenomic analyses revealed differentially abundant Kyoto Encyclopedia of Genes and Genomes (KEGG) Orthology (KO) categories in the almond (n = 54), broccoli (n = 2,474), and walnut (n = 732) groups. Using differentially abundant KOs in a random forest model led to prediction accuracies of 80% (almond), 87% (broccoli), and 86% (walnut) compared to their respective controls. A mixed-food random forest model revealed 81% prediction accuracy in differentiating these three foods from one another (2)(Chapter 4). Third, using 96 fecal metabolites, these six foods were differentiated from their respective control groups with prediction accuracies between 47%-89% and almond intake from walnut intake with 91% accuracy (3)(Chapter 5). Finally, our multi-omic analyses revealed 1,356 microbial species, 14,109 KOs, and 96 metabolites that achieved accuracies of 79% (almond), 87% (broccoli), and 94% (walnut) in a random forest model (Chapter 6). This effort provides groundwork for establishing non-invasive, fecal biomarkers of food intake, which may eventually be used as fidelity measures for dietary compliance and the development of personalized diet-microbiota tailored therapies for disease treatment and prevention.

Graduation Semester

2023-05

Type of Resource

Thesis

Handle URL

https://hdl.handle.net/2142/120548

Copyright and License Information

Copyright 2023 Leila Marie Shinn

Owning Collections