Engineering Multi-Protein Assemblies with Machine Learning and Experimental Optimization

Author: Dev, Ishaan Joshan

Year: 2027

Degree: Dissertation (Ph.D.)

Advisor: Shapiro, Mikhail G.

Committee Members: Arnold, Frances Hamilton; Shan, Shu-ou; Thomson, Matthew; Shapiro, Mikhail G.

Option: Chemical Engineering

DOI: 10.7907/gnma-ay69

Abstract

Mammalian Acoustic Reporter Gene (mARG) expression systems are used to express gas vesicle (GV) acoustic contrast agents, enabling visualization of mammalian cell dynamics in living organisms using ultrasound. Further advancement of mARGs is encumbered by all-too-common problems in synthetic biology for mammalian systems; there are limited frameworks to compactly express multiple protein components in distinct stoichiometries in a manner amenable for gene delivery. Here, we present new methods of polycistronic expression in mammalian cells through exploitation of the natural scanning behavior of ribosomes. We term this system Stoichiometric Expression of mRNA Polycistrons from Eukaryotic Ribosomes (SEMPER). The tunability of SEMPER systems is greatly reduced when proteins contain methionines within their structure. Therefore, we go on to demonstrate how zero-shot predictions yielded by a set of protein language models and structure-based methods can be used to generate methionine-free proteins with minimal loss in activity. As demonstrated on mARGs, SEMPER combined with our zero-shot prediction workflow provides a new and efficient toolkit by which synthetic biologists can encode multi-protein expression systems for complex synthetic biology applications. We complement these advances by characterizing cellular responses to gas vesicle expression using RNA-seq and by developing new methods to screen cellular acoustic properties.

Files