A protein that appears to be involved in limiting the inflammatory response to influenza infections is uncovered in a study published in Science Signaling this week. Type I interferons (IFNs) are released by cells and trigger changes in gene expression necessary to combat viral infections. Precise control of the IFN response is vital since insufficient expression of IFN-stimulated genes can result in the failure to stop viral spread, while excessive activation of these genes can lead to harmful inflammation and other pathologies. In the new study, Duke University scientists performed a CRISPR screen to identify previously unknown regulators of the type I IFN response. They find an IFN-stimulated gene that encodes a transcription factor called ETV7, which suppresses the expression of other IFN-stimulated genes including ones particularly important for IFN-mediated control of influenza viruses in lung epithelial cells. The results suggest that regulatory factors such as ETV7 may regulate the balance between control of viral infection and the excessive inflammation associated with severe viral disease.
A neural network-based method for protein modeling with accuracy near that of DeepMind's AlphaFold2 is reported in Science this week. At a recent CASP14 protein structure prediction assessment conference, AlphaFold — which was developed by Alphabet's DeepMind Technologies subsidiary — demonstrated remarkably accurate predictions, raising questions as to whether such accuracy could be achieved outside of a world-leading deep-learning company. In their paper, a team led by University of Washington scientists present RoseTTAFold, an AI network that produces structure predictions nearly as accurately as AlphaFold2 but requires a fraction of the computational processing power and time. The three-track network enables the rapid solution of challenging X-ray crystallography and cryo-EM structure modeling problems and provides insights into the functions of proteins of currently unknown structure, its developers write. It also enables rapid generation of accurate protein-protein complex models from sequence information alone compared with traditional approaches that require modeling of individual subunits followed by docking. RoseTTAFold is being made freely available to the scientific community.