Deep learning drives insight into protein-protein interactions
(Nanowerk News) Sequencing of proteins allows scientists to identify the amino acids in proteins. These amino acids determine the shape and function of proteins. AlphaFold 2 DeepMind is an artificial intelligence system originally designed to predict the shape of a single protein sequence.
In this study, scientists used AlphaFold 2 to develop a powerful deep learning approach to predict and model multi-protein interactions. His approach, AF2Complex, yields structural models that are much more accurate than previous methods for modeling protein complexes. AF2Complex can even predict novel protein-protein interactions.
As a proof of concept, the researchers used the AF2Complex to virtually screen proteins in the pathways that make the outer membrane in E. coli. This led to the discovery of unexpected protein-protein interactions.
Protein-protein interactions are essential for life. AF2Complex provides a powerful computational approach to detect and model these interactions. This approach can be applied to all complement proteins in cells. As a large-scale proof of concept, researchers use AF2Complex to examine an essential E. coli road. The work showcases the promise of deep learning-based research strategies for studying biological systems. It can help researchers understand many other biological systems by discovering new protein-protein interactions and offering high-quality predictions of their complex structures.
Life depends on molecular machinery made of proteins that interact with each other to form functional complexes. Researchers need accurate descriptions of protein-protein interactions to understand molecular biosystems, but obtaining such descriptions is challenging, especially for theoretical approaches. Up to now, protein-protein interactions have mainly been discovered and characterized by experimental approaches.
Generalizing AlphaFold 2, a powerful deep learning algorithm for predicting protein structure from sequences, researchers at the Georgia Institute of Technology and Oak Ridge National Laboratory proposed a computational approach, AF2Complex, to not only predict the atomic structure model of interacting proteins, but also to predict whether many proteins interact, even if they undergo transient interactions that are difficult to capture experimentally.
Scientists know that elusive examples occur in bacterial pathways that aid in the translocation and folding of outer membrane proteins. The team screened AF2Complex-activated virtual protein-protein interactions for several key proteins of this pathway against approximately 1,500 proteins in the cell envelope. E. coli.
The study used the Summit supercomputer at Oak Ridge National Laboratory. Among the most confident hits, the researchers identified known interaction partners and highly confident unexpected hits with implications for outer membrane biogenesis pathways. The resulting supercomplex structure model reveals several conformations that explain previous experimental observations and inspire new mechanistic hypotheses to understand how outer membrane proteins are made.
Publication
Gao, M., et al., AF2Complex predicts direct physical interactions in multimeric proteins by deep learning. Nature Communications 13, 1744 (2022). (DOI:10.1038/s41467-022-29394-2)
Gao, M., Nakajima An, D., & Skolnick, J. Deep learning-driven insights into superprotein complexes for outer membrane protein biogenesis in bacteria. eLife 11, e82885 (2022). (DOI:10.7554/eLife.82885)
