Towards sustainable biomolecule production: computational approaches to accelerate genetic tool development for engineering metabolism in microorganisms

Abstract

Globally, human societies are consuming finite resources at unsustainable rates. Transitioning away from our dependencies on non-renewable resources and towards cyclical production of everyday materials is critical for mitigating our escalating impact on climate change and securing longer term economic stability. A promising alternative to sourcing many materials is via metabolic engineering: a field that aims to engineer microorganisms into biological factories that convert renewable feedstocks into valuable molecules (i.e., jet fuel, medicine, bioplastics). In order for metabolic engineering solutions to be economically viable, microorganism factories must be optimized to produce target molecules quickly and at high yields. Such optimization requires an understanding of the complex genetic grammar that controls gene expression within a host microbe as well as genetic tools with which to manipulate it. While extensive genetic toolkits have been developed for model systems like S. cerevisiae and E. coli, many non-model organisms lack tools with which to effectively engineer them. This dissertation explores computational approaches for developing genetic tools in non-model microbes, using the methanotroph Methylotuvimicrobium buryatense as an example. First, we discuss a framework that leverages RNA-seq datasets to predict constitutive, strong promoters, which we developed into a suite of expression tools in M. buryatense. Next, we use unsupervised machine learning methods to identify 43 independently modulated groups of co-expressed genes (iModulons); interactively explorable visualizations of these data facilitated a deeper characterization of M. buryatense expression modules across diverse growth conditions and a proposed set of gene candidates for functional validation via mutation experiments. Finally, we investigate the potential of deep learning models to predict gene expression behavior directly from M. buryatense promoter sequence regions and probe the performance limits of common model architectures in varied genomic contexts and data-limited regimes. This work contributes to a broader understanding of how computational techniques can be used to model the effects of biological sequences on gene expression outcomes and describes scenarios in which these techniques are more limited. Such guidance will enhance our collective ability to use computational approaches for genetic tool development in non-model microbes and accelerate metabolic engineering solutions that can shift humanity towards a more sustainable relationship with our planet’s finite resources.