Tuesday, December 3, 2024
12/3/2024
Project Summary/Abstract Proteins are critical to all biological processes and designed proteins have important applications in research and medicine. This project focuses on the development and application of computational methods for protein design. Recently, there has been tremendous progress in the use of deep learning (DL) for protein modeling. Neural networks have been trained for structure prediction and sequence design that outperform more traditional molecular mechanics-based models in a variety of benchmarks. The exciting challenge before the field now is to discover how to extend these methods toward the design of the diverse functionalities and structural features evident in naturally occurring proteins. In preliminary studies, we developed a protein design pipeline, called EvoPro, that uses iterative rounds of DL-based structure prediction and sequence design to optimize a population of sequences for a user-specified design goal. As an initial test, EvoPro was used to create de novo proteins that bind to a specified surface patch on another protein. Without any affinity maturation, EvoPro generated several proteins that bind to the target with KDs less than 150 nM. We now aim to test this pipeline on a diverse set of protein targets, choosing cases where protein binders may be useful as biosensors or therapeutics. Allostery is a fundamental process used throughout biology for regulating protein function, and because of technical challenges is an understudied area in protein design. We will use EvoPro to design protein interactions where the binding of protein A to protein B induces a conformational change in protein B. To enable this goal, EvoPro will make use of DL models for sequence design that can optimize an amino acid sequence to be compatible with multiple conformations. Progress in this area will enable the design of protein switches that respond to positive and negative regulators. Explicit negative design will also be incorporated into the sequence design models to redesign the specificity of protein-protein interactions and enable the formation of novel protein assemblies. To improve DL models for sequence design, we are testing graph neural networks that include explicit representations of side chain atoms during message passing. These models are being trained to simultaneously predict side chain coordinates and optimize the protein sequence. By pursuing these projects, we will expand the capabilities of computational protein design and create molecules that can be used to understand or treat disease.
HHS’ Tracking Accountability in Government Grants System (TAGGS) website provides access to detailed descriptions of grants, loans, aggregated direct payments and other types of financial assistance awarded by the HHS Operating Divisions (OPDIVs) and the Office of the Secretary Staff Divisions (STAFFDIVs).
