Revised codes for no toxin chains, Manual epitope design

[Hyeongsun]

Excellent work, I tried to generate antibody for COVID NP protein.

So, I revised the code as below;

if name == 'main':
ckpt = './checkpoints/cdrh3_design.ckpt'
root_dir = './demos/data'
pdbs = [os.path.join(root_dir, '6wzo.pdb') for _ in range(4)]
toxin_chains = []
remove_chains = None
receptor_chains = ["A", "B", "C", "D"]
epitope_defs = [os.path.join(root_dir, c + '_epitope.json') for c in receptor_chains]
identifiers = [f'{c}_antibody' for c in receptor_chains]

I manually design epitope.json file as;

A_epitope.json
[
["A", [299, ""]],
["A", [300, ""]],
["A", [302, ""]],
["A", [303, ""]],
["A", [305, ""]],
["A", [306, ""]],
["A", [345, ""]],
["A", [347, ""]],
["A", [348, ""]],
["A", [349, ""]]
]
However, I encounter the messeges as;

/home/jhs9301/anaconda3/envs/dyMEAN/lib/python3.8/site-packages/numpy/core/fromnumeric.py:3464: RuntimeWarning: Mean of empty slice.
return _methods._mean(a, axis=axis, dtype=dtype,
/home/jhs9301/anaconda3/envs/dyMEAN/lib/python3.8/site-packages/numpy/core/_methods.py:184: RuntimeWarning: invalid value encountered in divide
ret = um.true_divide(
/mnt/c/Users/jhs93/Downloads/dyMEAN-main/dyMEAN-main/data/dataset.py:63: RuntimeWarning: invalid value encountered in cast
X[0] = center # set center

I guess the invalid values were obtained in Epitope data.

Epitope data: {'X': array([[[-9223372036854775808, -9223372036854775808,
-9223372036854775808],
[-9223372036854775808, -9223372036854775808,
-9223372036854775808],
[-9223372036854775808, -9223372036854775808,
-9223372036854775808],
[-9223372036854775808, -9223372036854775808,
-9223372036854775808],
[-9223372036854775808, -9223372036854775808,
-9223372036854775808],
[-9223372036854775808, -9223372036854775808,
-9223372036854775808],
[-9223372036854775808, -9223372036854775808,
-9223372036854775808],
[-9223372036854775808, -9223372036854775808,
-9223372036854775808],
[-9223372036854775808, -9223372036854775808,
-9223372036854775808],
[-9223372036854775808, -9223372036854775808,
-9223372036854775808],
[-9223372036854775808, -9223372036854775808,
-9223372036854775808],
[-9223372036854775808, -9223372036854775808,
-9223372036854775808],
[-9223372036854775808, -9223372036854775808,
-9223372036854775808],
[-9223372036854775808, -9223372036854775808,
-9223372036854775808]]]), 'S': [22], 'residue_pos': [0], 'xloss_mask': [[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]}

Then, How can i fix the problem? please let me know.

[kxz18]

Hi, thanks for your interest in our work. Could you try replacing the insert code with a space instead of an empty string? The code base is using biopython to parse the PDB file, and the biopython package uses a space to denote an empty insert code. This mismatch might be the cause of the problem.

[Hyeongsun]

Thanks! It was fixed. there is one more question. Is there any way to obtain the complex structure of an antibody sequence that I already know and an antigen? I wanna optimize the already developed antibody but there is no complex structure.

[kxz18]

Hi, I've updated the API for structure prediction here. Corresponding explanations are also updated in README. You can have a try!

[YaweiYu]

Excellent work! I have confusion about the structure prediction tool. Predicte the complex need to provide antigen structure, antibody sequence and epitope definition, but how can we get epitope when there is no complex structure? Looking forward to hearing from you!

[kxz18]

@YaweiYu Hi, the structure prediction tool is mostly a local docking & structure prediction tool, which assumes that the user already has some knowledge on the binding site of the antibody, just like popular docking softwares for small molecules (e.g. glide, vina). For example, if you know that the antibody is blocking some protein-protein interactions, than it is highly possible that the antibody is approximately binding to the interface of the PPI. Under such circumstances, you can use the protein-protein complex to extract the epitope definition for local docking & structure prediction.

Global docking/cofolding of antibodies to proteins is still a challenging problem, on which most existing models cannot produce satisfactory performance, and the most appealing one might still be Alphafold 3.

[github-actions]

This issue is stale because it has been open for 30 days with no activity.