Communications of the ACM
AF2Complex: Researchers Leverage Deep Learning to Predict Physical Interactions of Protein Complexes
A three-dimensional rendering of the structures of two protein complexes, predicted from protein sequences by AF2Complex.
Credit: Mu Gao
Researchers at the Georgia Institute of Technology (Georgia Tech) and the U.S. Department of Energy's Oak Ridge National Laboratory have built a deep learning model to predict the biologically active structure of proteins and protein complexes.
The model is based on artificial intelligence lab DeepMind's AlphaFold 2 neural network, which is designed to extrapolate the three-dimensional structure of a single protein from its amino acid sequence.
Rather than plugging in the features of an individual protein sequence into AlphaFold 2 as original designed, the researchers blended the input features of multiple sequences, which in combination with new metrics to evaluate protein interaction strength produced the AF2Complex algorithm.
AF2Complex identified which of 500-plus protein pairs were able to form a complex, and it outperformed traditional docking methods and AlphaFold 2 in recognizing arbitrary pairs known to experimentally interact.
From Georgia Institute of Technology
View Full Article
Abstracts Copyright © 2022 SmithBucklin, Washington, DC, USA
No entries found