Introductions

Rosetta's Ab initio method of predicting protein structure can more accurately verify the de novo protein sequence and structure matching. The free energy surface of protein folding is generally considered to be funnel-shaped, and the natural structure of protein is at the point of global energy minimum, which has a significant energy difference from the conformation of other folding intermediate states. Therefore, if the artificially designed protein has a similar energy funnel, the rationality of the designed structure can be evaluated. However, Ab initio has a huge amount of calculation, which severely limits its application in the process of de novo design. Biased forward folding simulations is a very economical simulation method, which can quickly identify which design structures are easier to fold into the designed target state during Ab initio simulation. The Biased forward folding scheme allows us to verify the foldability of thousands of structures during the design process. In order to exclude those structures and sequences that may not be expressed in the experiment.

1. Prepare Fragment file

Biased forward folding is a simplified version of Ab initio folding. First of all, we need to obtain the fragment set for re-folding simulation. There are many ways to generate fragments. You can use the FragmentPicker that comes with Rosetta, or you can use other methods. The specification must conform to Rosetta's Fragment format.

Notes
Add FragmentCrmsd weight when creating wghts weight file
Use -frags::describe_fragments to output fragment specific information files, such as files ending with .fsc. Documented in this file
Use -in::file::s to enter the protein structure generated by de novo

2. Separate the Fragment subset that is most similar to the structure

The github source of Biased forward folding here has open-source code, but some apps have been removed in Rosetta's public release. Therefore, we provide the latest modified version, the usage method is as follows:
python lowrms_frags_topN.py -frag_qual frags.fsc.3mers -ntop 3 -fullmer frags.3mers -out top3.3mers
python lowrms_frags_topN.py -frag_qual frags.fsc.9mers -ntop 3 -fullmer frags.9mers -out top3.9mers
ag_qual: Enter the above fragment scoring file
ntop: Pick the top X fragments for simulation
fullmer: Fragment original file
out: output new Fragment subset file

It is worth noting that here we need to generate 9- and 3-mer fragment files at the same time. By default, it is recommended to use the first 3 fragments with the lowest RMSD of the Cα atom for fast simulation.

3. Perform Ab initio folding simulation

With the new Fragment subset, we can use the standard ab initio structure prediction app for structural simulation.

It is recommended to sample 30-50 simulated trajectories to evaluate whether our target conformation appears in ab initio folding.

Method:
AbinitioRelax.default.macosclangrelease -in:file:native 2jsvX.pdb -in:file:fasta 2jsvX.fasta -in:file:frag3 top3.3mers -in:file:frag9 top3.9mers -abinitio:relax -relax::fast -abinitio::increase_cycles 10 -ex1 -ex2aro -abinitio::use_filters false -abinitio::rsd_wt_loop 0.5 -abinitio::rsd_wt_helix 0.5 -abinitio::rg_reweight 0.5 -out:pdb -nstruct 30

Next, you can continue to analyze whether the energy funnel shape or whether the output target PDB and the designed structure RMSD are less than a certain cut-off value to judge the rationality of the design, etc.