If redesigned_residues_multi is supplied, the list of residues to redesign in not correct

[oteri]

Hello,
thank you for the hard work you have been doing.
I think I spotted a bug at this line:

LigandMPNN/run.py

Line 107 in b161414

redesigned_residues_multi = json.load(fh)

I think, after the files is loaded, the list of residues must be split like it is done at line 109.

This is how I fixed the bug ( don't know if it is right, though):

    if args.redesigned_residues_multi:
        with open(args.redesigned_residues_multi, "r") as fh:
            redesigned_residues_multi = json.load(fh)
            redesigned_residues_multi = { pdb:redesigned_residues.split() for pdb,redesigned_residues in redesigned_residues_multi.items() }

Thank you again

[geraseva]

Same problem

[geraseva]

This is how I fixed the bug ( don't know if it is right, though):

    if args.redesigned_residues_multi:
        with open(args.redesigned_residues_multi, "r") as fh:
            redesigned_residues_multi = json.load(fh)
            redesigned_residues_multi = { pdb:redesigned_residues.split() for pdb,redesigned_residues in redesigned_residues_multi.items() }

Your fix worked for me, thank you!

[chubbj]

This is how I fixed the bug ( don't know if it is right, though):

    if args.redesigned_residues_multi:
        with open(args.redesigned_residues_multi, "r") as fh:
            redesigned_residues_multi = json.load(fh)
            redesigned_residues_multi = { pdb:redesigned_residues.split() for pdb,redesigned_residues in redesigned_residues_multi.items() }

This fixed also worked for me, thanks.

[annie393]

In run.py, the residues to be fixed are defined as at line 193 and line 244:

# line 193
fixed_residues = fixed_residues_multi[os.path.basename(pdb).replace('.pdb', '')]
# line 244
fixed_positions = torch.tensor(
            [int(item not in fixed_residues) for item in encoded_residues],
            device=device,
        )

with input in the example as:
--fixed_residues "C1 C2 C3 C4 C5 C6 C7 C8 C9 C10"

In that case, if C72 is included in fixed_residues, then C7 is included in fixed_residues. Maybe a line should be added after line 193:
fixed_residues = fixed_residues.split(' ')

I am not sure whether this will be helpful for this problem.

[Gabriel-QIN]

❓ Unexpected High Mutation Count with --ligand_mpnn_use_atom_context 1 and Recovery ~56%

Hi, thank you for this great tool. I have encountered a confusing issue when using ligandMPNN with ligand context enabled.

🧪 Setup
Redesigned residues (explicitly defined via --redesigned_residues):

A237 A241 A244 A276 A281 A282 A283 A284 A285 A288 A292 A293 A294 A296 A299 A301 A339 A340 A341 A363 A364 A365 A366 A367 A368 A369 A386 A387 A388 A398 A401 A416 A437 A439 A440 A441 A442 A444 A445 A465 A466 A482 A483 A484 A485 A486 A489 A490 A492 A493 A499 A504 A505 A507 A508 A509 A510 A511 A513 A537 A538 A540 A541 A542 A543
Total redesigned residues: 65

Ligand residues: 52

I used:

--ligand_mpnn_use_atom_context 1 \
--redesigned_residues "..."  \
--ligand_cutoff_distance 8.0 \
--use_ligand_context True \
--batch_size 100 \
--number_of_batches 10

Although I manually specified only 65 residues for design, the output sequences often contain >50 mutations compared to the wild-type, and the recovery is around 56%.

This seems inconsistent — I expected recovery to be around (1 - 50/65) ≈ 23%, or the mutation count to be closer to 65 * (1 - 0.56) ≈ 29. But neither aligns.

My Questions:

1. Does setting --ligand_mpnn_use_atom_context 1 automatically include more residues around the ligand for redesign, even when --redesigned_residues is explicitly set?

2. Is it possible that ligand context overrides or expands the --redesigned_residues list behind the scenes?

3. Could you clarify the interaction between --ligand_mpnn_use_atom_context and --redesigned_residues?

4. Are there any logging options to confirm exactly which residues were redesigned in each sequence?

The following snippet from my run.py shows how the --ligand_mpnn_use_atom_context flag and related fixed residues are parsed and handled:

### line 126-134
if args.fixed_residues_multi:
    with open(args.fixed_residues_multi, "r") as fh:
        fixed_residues_multi = json.load(fh)
        fixed_residues_multi = {key:value.split() for key,value in fixed_residues_multi.items()}
else:
    fixed_residues = [item for item in args.fixed_residues.split()]
    fixed_residues_multi = {}
    for pdb in pdb_paths:
        fixed_residues_multi[pdb] = fixed_residues