摘要：SummaryWith the advent of new artificial intelligence and machine learning algorithms, predictive modeling can, in some cases, produce structures on par with experimental methods. The combination of predictive modeling and experimental structure determination by electron cryomicroscopy (cryoEM) offers a tantalizing approach for producing robust atomic models of macromolecular assemblies. Here, we apply AlphaFold2 to a set of community standard data sets and compare the results with the corresponding reference maps and models. Moreover, we present three unique case studies from previously determined cryoEM density maps of viruses. Our results show that AlphaFold2 can not only produce reasonably accurate models for analysis and additional hypotheses testing, but can also potentially yield incorrect structures if not properly validated with experimental data. Whereas we outline numerous shortcomings and potential pitfalls of predictive modeling, the obvious synergy between predictive modeling and cryoEM will undoubtedly result in new computational modeling tools.Graphical abstractDisplay OmittedHighlights•AlphaFold2 can be combined with cryoEM density maps for macromolecular modeling•Even at lower resolutions, AlphaFold2 and cryoEM can generate testable models•Predictive modeling must be validated by experimental data for reliabilityComputational molecular modelling; Biochemistry; Structural biology
