Evolution as a design tool to inform biomolecular engineering

Enzyme biotechnology is a critical component of technologies needed for increased sustainable materials processing. Along with the ability to rapidly synthesize proteins through fermentation, there is a need to be able to alter enzyme functionality in specific ways to suit the desired application. For instance, industrial enzymes with increased stability at higher temperatures or altered pH optima can improve productivity in large-scale bioreactors through improved catalytic rates and lowered costs due to cooling and reduced contamination. This research aims to provide the scientific community with a suite of design tools and methodologies for protein engineering experiments and research. Specifically, these methods utilize a foundation of concepts from molecular evolution to provide insight into the innovation process in molecular engineering. Methods developed in this study aim to increase thermal stability of an enzyme by engineering disulfide bonds and electrostatic salt bridges on the surface of the enzyme. Two methods to assist disulfide engineering were developed- 1) A neural network model to predict disulfide bonds within existing structures from mutual information and a continuous distributed representation of protein sequence. 2) A methodology incorporating statistics on the structural information of disulfide bonds in conjunction with evolutionary patterns to rank order specific design choices. A similar approach was developed for engineering salt bridges. The methodology for developing geometric constraints and utilizing evolutionary patterns to engineer salt bridges was validated with experiments on 1,4 -glucan branching enzymes by collaborators. The neural network model achieves state of the art accuracy (80%) and in addition, the impact of the protein sequence representation, mutual information and cysteine separation distance on performance of the model were analysed. in a particular disulfide engineering experiment. The trained long short term memory (LSTM) neural network model also serves as a model of disulfide bond sequence motifs so as to develop an understanding of the constraints for disulfide bond formation. The methodology using statistical constraints on disulfide bonds was prototyped as a PyMOL script that identifies potential pairs of residues on the surface of an enzyme for modification to disulfide-capable cysteine residues. This method suggests 85% more stabilising mutations out of 17% fewer suggestions according to evaluations by short Molecular Dynamics simulations using FoldX