Scientists have waited months for access to high-accuracy protein structure prediction since DeepMind presented remarkable progress in this area at the 2020 Critical Assessment of Structure Prediction, or CASP14, conference. The wait is now over.
Researchers at the Institute for Protein Design at the University of Washington School of Medicine in Seattle have largely recreated the performance achieved by DeepMind on this important task. These results were published online July 15 by the journal Science.
Unlike DeepMind, the UW Medicine team has already made their method, dubbed RoseTTAFold, freely available. Scientists from around the world are now using it to build protein models to accelerate their own research. Soon after its recent upload, the program was downloaded from GitHub by over 140 independent research teams.
Proteins consist of strings of amino acids that fold up into intricate microscopic shapes. These unique shapes in turn give rise to nearly every chemical process inside living organisms. By better understanding protein shapes, scientists can speed up the development of new treatments for cancer, COVID-19, and thousands of other medical disorders.
In the new study, a team of computational biologists led by Baker developed a software tool called RoseTTAFold that uses deep learning to quickly and accurately predict protein structures based on limited information. Without the aid of such software, it can take years of laboratory work to determine the structure of just one protein.
RoseTTAFold, on the other hand, can reliably compute a protein structure in as little as 10 minutes on a single gaming computer. The team used RoseTTAFold to compute hundreds of new protein structures, including many poorly understood proteins from the human genome. They also generated structures directly relevant to human health, including for proteins associated with problematic lipid metabolism, inflammation disorders, and cancer cell growth.
RoseTTAFold is a “three-track” neural network; meaning, it simultaneously considers patterns in protein sequences, how a protein’s amino acids interact with one another, and a protein’s possible three-dimensional structure. In this architecture, one-, two-, and three-dimensional information flows back and forth, allowing the network to collectively reason about the relationship between a protein’s chemical parts and its folded structure.
