Protein Structures Can Be Predicted in 10 Minutes Using New Artificial Intelligence Software

Anyone now has access to accurate protein structure prediction.
Since DeepMind’s excellent achievement in this sector was announced at the 2020 Critical Assessment of Structure Prediction, or CASP14, conference, scientists have waited months for access to very precise protein structure prediction. Finally, the wait is over.

DeepMind’s performance on this key task has been largely replicated by researchers at the University of Washington School of Medicine’s Institute for Protein Design in Seattle. The journal Science published these findings online on July 15, 2021.
Unlike DeepMind, the technique used by the UW Medicine team, dubbed RoseTTAFold, is open source. Scientists from all across the world are already using it to rapidly generate protein models for their own studies. Since July, the application has been downloaded by over 140 separate research teams via GitHub.

Proteins are composed of amino acid chains that fold into complicated, microscopic structures. These many forms, in turn, are responsible for virtually every chemical process occurring within living organisms. By gaining a better understanding of protein structures, scientists can accelerate the development of innovative therapies for cancer, COVID-19, and thousands of other disorders. The Institute for Protein Design has had a busy year, producing COVID-19 medications and vaccines and advancing them into clinical trials, as well as inventing RoseTTAFold, a high-precision protein structure prediction tool. “I am ecstatic that the scientific community is already utilising the RoseTTAFold server to address outstanding biological problems,” said senior author David Baker, a biochemistry professor at the University of Washington School of Medicine, a Howard Hughes Medical Institute investigator, and director of the Institute for Protein Design.

For the newest work, a team of computational biologists led by Baker developed the RoseTTAFold software tool. Deep learning is used to fast and reliably predict the structures of proteins using sparse information. Without such software, it can take years of laboratory work to determine the structure of a single protein.
RoseTTAFold, on the other hand, can compute the structure of a protein in as little as ten minutes using a single gaming computer.

The scientists utilised RoseTTAFold to compute hundreds of unique protein structures, including numerous previously undiscovered proteins from the human genome. Additionally, they developed structures associated with human health, such as those for proteins implicated in improper cholesterol metabolism, inflammatory diseases, and cancer cell proliferation. Additionally, they demonstrate how RoseTTAFold can be used to rapidly generate models of complex biological assemblies in a fifth of the time necessary previously.

RoseTTAFold is a “three-track” neural network, which means it simultaneously examines patterns in protein sequences, the interactions of amino acids, and a protein’s possible three-dimensional structure. This architecture allows for the exchange of one-, two-, and three-dimensional information, allowing the network to reason collectively about the relationship between a protein’s chemical contents and its folded shape.

“We hope that this new tool will continue to benefit the broader research community,” said Minkyung Baek, a postdoctoral scholar at the University of Washington School of Medicine who spearheaded the initiative in the Baker laboratory.

Reference
Baek, M., DiMaio, F., Anishchenko, I., Dauparas, J., Ovchinnikov, S., Lee, G. R., … & Baker, D. (2021). Accurate prediction of protein structures and interactions using a three-track neural network. Science.