NovaFold AI wizard: Options

When you predict a protein structure with NovaFold AI, the second wizard screen is Options. During the prediction step, options specified in this screen are applied to all jobs listed in the Sequences screen.

Edit or enter information in this wizard screen as desired. Or you can accept the defaults as they are, and simply press Next to continue to the Submit screen.

• Size of genetic database – NovaFold AI constructs a deep multiple sequence alignment as input for the structure prediction process. Use the drop-down menu to choose either:
  
  
  • Extra large – To search through all 2 billion protein sequences in the full BFD.
    
    
  • Large – To search a reduced version of the BFD prepared by the AlphaFold group.

• Trained network configuration – Use the drop-down menu to choose between:
  
  
  • Standard – The default model.
    
    
  • Extra ensembling – The model used during the CASP14 competition. It requires 8x more computation but yields only a minor gain in modeling accuracy.
    
    
  • Extra metrics – A model tuned to provide a pairwise confidence measure (predicted C-alpha differences; also used by NovaFold/I-TASSER). This model is less accurate than the standard model.

• Relax energy of models – Check the box if you wish to minimize the energy of the prediction. Leave the box unchecked to use the raw coordinates from the AI model.

• Use templates published until – Today’s date is shown by default. To limit templates to those published before an earlier date, specify the date by using the up/down triangles or by clicking on the Choose a date tool (). This option was designed for DNASTAR’s internal benchmarking purposes. It can be used to reproduce the CASP14 experiment with a contemporary set of the Protein Data Bank.

Click Next to proceed to the Submit screen, or Back to return to the Sequences screen.