Communications of the ACM
Deep Learning for New Protein Design
Deep learning methods augmented existing energy-based physical models in computational protein design.
Credit: National Cancer Institute
Scientists at the University of Washington (UW) and Belgium's Ghent University enhanced current energy-based physical models in de novo computational protein design using deep learning techniques.
The researchers incorporated DeepMind's AlphaFold 2 and the UW-developed RoseTTA fold software into the deep learning-augmented de novo protein binder design protocol. They ran 6 million interactions between potentially bound protein structures in parallel on the Texas Advanced Computing Center's Frontera supercomputer and used UW's ProteinMPNN software to produce protein-sequence neural networks over 200 times faster than the previous best software.
Outcomes indicated the designed structures bind to target proteins 10 times faster, though UW's Brian Coventry said they must boost their speed by another three orders of magnitude. The work is described in Nature Communications.
From Texas Advanced Computing Center
View Full Article
Abstracts Copyright © 2023 SmithBucklin, Washington, D.C., USA
No entries found