News Feature Correction for “Researchers turn to deep learning to decode protein structures,” by Stephen Ornes, which published March 2, 2022; 10.1073/pnas.2202107119 (Proc. Natl. Acad. Sci. U.S.A. 119, e2202107119).
The author notes that on page 3, right column, first full paragraph, line 3, “…the algorithms have to train on big datasets of high-resolution crystal structures generated by sophisticated technologies such as NMR spectroscopy or cryogenic electron microscopy (cryo-EM). 'They go hand in hand,' he says, noting that AI models are nearly as good at predicting structure as those advanced experimental methods.” should instead appear as “…the algorithms have to train on big datasets of high-resolution structures generated by sophisticated technologies such as X-ray crystallography, nuclear magnetic resonance (NMR) spectroscopy, or cryogenic electron microscopy (cryo-EM). 'They go hand in hand,' he says, noting that AI models may predict structures that can’t be determined by advanced experimental methods.” The online version has been corrected.
The author notes that on page 3, left column, first full paragraph, line 2, “Until 2016, entrants rarely achieved higher than 20% accuracy. Then AI entered the scene. That year, during CASP12, computational biologist David Jones from the University College London, UK, more than doubled previous accuracy levels using a model powered by deep learning algorithms. In the wake of his work, neural networks took the CASP community—and the broader field of structural biology—by storm.” should instead appear as “Until 2016, entrants rarely achieved higher than 20% accuracy in predicting contacts between amino acids in the test proteins. That year, during CASP12, computational biologist David Jones from the University College London, UK, achieved an accuracy of about 27%. Then AI entered the scene: In CASP13, a group led by Yang Zhang at the University of Michigan and Jinbo Xu at the Toyota Technical Institute at Chicago used deep learning to nearly double that number. In the wake of these efforts, neural networks took the CASP community—and the broader field of structural biology—by storm.” The online version has been corrected.
The author notes that the legend for Fig. 2 appeared incorrectly. The figure and its corrected legend appear below. The online version has been corrected.
The revolution in structural biology isn’t attributable to AI alone; the algorithms have to train on big datasets of high-resolution structures generated by technologies such as X-ray crystallography, NMR spectroscopy or cryogenic electron microscopy (cryo-EM), which produced the above image of a protein complex called β-galactosidase. Image credit: Veronica Falconieri and Sriram Subramaniam (National Cancer Institute, Bethesda, MD).