Computational Design of Sequence-Specific DNA-Binding Proteins

Abstract

The interactions between proteins and DNA are fundamental to converting genetic information stored in DNA into the systems of life, which are often driven by proteins, in a constant feedback loop. A deeper understanding of how these protein-DNA interactions work would elucidate the mechanisms of cell states in which improper regulation turns into disease and could translate into new tools for studying and manipulating these systems, paving the way to target and cure genetic diseases. In this dissertation, I present a set of computational methods to design new protein-DNA interactions. These methods include 1) a pipeline to design nature-inspired protein scaffolds using both structure-based filters and resampling techniques to produce DNA-binding proteins; 2) a message-passing neural network trained to predict the sequence of an amino acid at a position given nucleic acid or general ligand context; and 3) a diffusion-based model to generate de novo protein backbones to bind to DNA. These tools have enabled the design of several orthogonal DNA-binding proteins which bind their targets with biologically relevant affinities. These interactions were validated through extensive protein- and DNA-side mutations, screens against on- and off-target sequences, and, in one case, a crystallographic structure demonstrating the accuracy of our designed complex. Additionally, I discuss co-leading an undergraduate research training program called JUPITER, where these methods were taught to University of Washington students who were learning about protein design for the very first time. JUPITER students not only designed their own successful binders but also contributed meaningfully to functionalizing designed proteins as repressors of transcription in E. coli. Altogether, these improvements to the repertoire of computational protein design will enable future researchers to target new types of molecules in the pursuit of synthetic biology and medical innovation.