This Deep Learning Model ‘RoseTTAFold’ Can Compute A Protein Structure In As Little As Ten Minutes On A Single Gaming Computer

1246
Source: https://www.ipd.uw.edu/2021/07/rosettafold-accurate-protein-structure-prediction-accessible-to-all/

In a recent study, researchers from the University of Washington have developed an innovative model that is capable of using gaming computers to calculate protein structures within 10 minutes. This research will likely accelerate drug development and provide solutions for cancer and other diseases.

Researchers from DeepMind have created a revolutionary piece of software that can predict the structure of proteins without relying on knowledge about how their amino acid sequence corresponds to 3D shape. This is an unprecedented breakthrough in research, and should lead to more advances as related protein studies progress.

Now, researchers at the University of Washington have developed a powerful three-track neural network, RoseTTAFold model, that is capable of considering protein sequence patterns, amino acid interactions and 3D structures. The research team used discrete fragments to train this model which had 260 unique elements in it. They used Pytorch framework to infer the protein’s chemical composition and folding structure.

https://science.sciencemag.org/content/early/2021/07/19/science.abj8754

DeepMind used powerful GPU operations for several days before it was able to make a single prediction. Now, the end-to-end version of RoseTTAFold on RTX 2080 GPUs only takes about 10 minutes to generate skeleton coordinates for proteins with less than 400 residues and can be much more accurate in its predictions as well. This software could not only help speed up protein synthesis but also use limited input or even predict complex substances composed of many proteins.

Paper: https://science.sciencemag.org/content/early/2021/07/19/science.abj8754

 RoseTTAFold Public Server: https://robetta.bakerlab.org/

Github: https://github.com/RosettaCommons/RoseTTAFold