Data Driven Design of Protein Function
Abstract
Our research first focused on a design task: designing mini-proteins that are stable, without requiring them to be functional. This work involved designing, testing, and learning from hundreds of thousands of mini-proteins, and resulted in machine learning (ML) models that are predictive of stability, as well as improvements to Rosetta's energy function. In the last two years, we have advanced to substantially more difficult tasks: designing proteins that are both stable and bind to a protein or DNA target. This work involved testing hundreds of thousands of binders, has resulted in dramatic increases to our success rates, and has provided several high-throughput datasets for future learning.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 24, 2022
- Accession Number
- AD1164648
Entities
People
- David Baker
- Hugh Haddox
- Ian C Haydon
- Kristina Herrera
- Lance Stewart
Organizations
- University of Washington