The first step is to fold the protein from the stretched-out conformation . This is done by the macro 'nmr_fold' using the NMRFolding experiment , the process is described in more detail there.

At this stage, speed is more important than accuracy, and the structures generated this way are not realistic proteins yet. But they have helices at the right spot and the peptide chain running in the right direction to quickly arrive at the correct solution during the following molecular dynamics refinement.

The number of structures generated in this stage is specified by the 'structures' parameter in 'nmr_setdefaults' and is equal to the final number of ensemble members.

If you are running Linux, you can of course also skip this step and use other programs like Concoord to fold the structures.