Doctoral Dissertations
Date of Award
12-2025
Degree Type
Dissertation
Degree Name
Doctor of Philosophy
Major
Life Sciences
Major Professor
Scott J. Emrich
Committee Members
Scott Emrich, Andrew Steen, Tongye Shen, Tatiana Vishnivetskaya
Abstract
Uncultured microbes constitute the vast majority of microbial diversity on Earth, presenting significant obstacles to understanding their ecological roles and physiological adaptations through traditional cultivation techniques. The advent of metagenome-assembled genomes (MAGs) has greatly expanded our insights into uncultured microbes, but major challenges remain in linking genomic traits to culturability and growth potential. This dissertation addresses a subset of these challenges by investigating genomic traits associated with culturability, exploring unexpected slow growth adaptation in cold environments and applying deep learning models to predict microbial growth rate directly from the genomic sequences. In the first part of this work, I analyzed 52,515 MAGs from the Genomes from Earth’s Microbiomes (GEM) catalog to examine functional novelty and culturability across phylogenetic lineages. Uncultured MAGs, particularly among Archaea, were found to encode significantly more divergent proteins. Using pathway enrichment analysis, LASSO regression, and permutation-based feature importance, I identified a core set of genomic traits predictive of culturability, suggesting that culturability is linked to conserved functional signatures across microbial lineages. Cultured MAGs were enriched in vitamin and cofactor biosynthesis, which were also strong predictors of microbial culturability. The second part of this study focuses on permafrost soils, where energy availability is chronically low. Here, I tested the hypothesis that thermophilic traits may confer an advantage to extremely slow-growing microbes in cold environments. Genomic indicators of thermophilicity were positively correlated with predicted doubling times in permafrost, revealing a counterintuitive pattern in which thermophilic traits appear to be adaptive for persistence under slow-growth conditions. This trend was reversed in seasonally thawed active layer soils, where periodic energy influx supports faster microbial growth. To improve prediction of microbial growth rate, I developed LGMrib, a transformer-based deep learning model fine-tuned on ribosomal protein-coding sequences. This model significantly outperforms traditional codon usage-based methods, particularly for slow-growing microbes. Applied across 25,000+ genomes, it reveals pervasive cultivation biases toward fast growers across diverse environments. Collectively, this work integrates computational and machine learning techniques to advance our understanding of microbial growth potential, culturability, and environmental adaptation, also offering frameworks for guiding cultivation strategies and predicting microbial traits from metagenomic data.
Recommended Citation
Oduwole, Iyanu, "Applying Bioinformatics and Machine Learning Techniques to Investigate the Cultivability and Slow Growth of Uncultured Microbes. " PhD diss., University of Tennessee, 2025.
https://trace.tennessee.edu/utk_graddiss/13626