Clarification Re: Motif Scaffolding and Binder Design

[amelie-iska]

I am attempting to perform motif scaffolding and binder design with RFD-AA similar to the way it was done in the original RFD. I ran into issues with the following function in idealize_backbone.py. The current implementation requires a ligand to be in the input. I commented the original lines in the code block and added some modifications to handle the case when there is no ligand (see below). Do my modifications seem reasonable, and is this use case supported?

def rewrite(path, outpath):
    with open(path, 'r') as fh:
        stream = [l for l in fh if "HETATM" in l or "CONECT" in l]

    ligands = get_ligands(stream)
    # indep = aa_model.make_indep(path, ','.join(ligands), center=False)
    if len(ligands) > 0: 
        indep = aa_model.make_indep(path, ','.join(ligands), center=False)
    elif len(ligands) == 0:
        indep = aa_model.make_indep(path, ligand=None, center=False)
    xyz = indep.xyz[~indep.is_sm]
    idx = indep.idx[~indep.is_sm]
    L = xyz.shape[0]
    ala_seq = torch.zeros((L,))
    xyz = rf2aa.util.idealize_reference_frame(ala_seq[None], xyz[None])[0]
    xyz_ideal = get_o(xyz, idx)
    indep.xyz[~indep.is_sm, :4] = xyz_ideal
    ligands = list(ligands)
    # assert len(ligands) == 1, f'Found >1 ligand: {ligands}'
    assert len(ligands) <= 1, f'Found >1 ligand: {ligands}'
    # indep.write_pdb(outpath, lig_name=ligands[0])
    if len(ligands) == 1:
        indep.write_pdb(outpath, lig_name=ligands[0])
    else:
        indep.write_pdb(outpath)

Another issue that I ran into was during binder design (PPI) with RFD-AA, the generated design seems to just be the same target protein given as input. The command I ran is:

python run_inference.py \
    diffuser.T=50 \
    inference.output_prefix=output/design_ppi/binder \
    inference.input_pdb=input/insulin_target.pdb \
    contigmap.contigs=[\'A1-150\',\'70-100\'] \
    ppi.hotspot_res=[A59,A83,A91] \
    inference.num_designs=1 \
    denoiser.noise_scale_ca=0 \
    denoiser.noise_scale_frame=0

Is this use case supported as well? If so, what might be the cause of the issue?