Predicting initial models with AlphaFold

AlphaFold neural network predicts the 3D coordinates of all heavy atoms for a given protein using solely the primary amino acid sequence as input [Jumper et al., 2021].

In this tutorial we show how to generate AlphaFold models of your sequence and rebuild them using the 3D density map. We use the Scipion framework and the approach described in the PHENIX web site and summarized as:

  • Get the initial AlphaFold model

  • Remove low-confidence residues of the model and break it into compact domains

  • Dock the domains obtained in the previous model processing step into your cryo-EM map or unit cell

  • Morph the docked fragments and rebuild the whole predicted model in the map density