ligandmpnn

1# LigandMPNN
2 
3LigandMPNN extends the ProteinMPNN graph with non-protein atoms — small
4molecules, nucleic acids, and metals are visible to the network — so it is the
5right inverse-folding tool whenever the design surface includes a bound ligand
6or cofactor that vanilla `proteinmpnn` would ignore. The same `run.py` is also
7the most convenient runner for the other MPNN families because, unlike the
8original ProteinMPNN script, it threads designs back onto the input structure
9and writes PDBs alongside the FASTA. Code and weights are MIT
10(github.com/dauparas/LigandMPNN). The model is small enough to run on CPU —
11for a handful of designs on one structure that is seconds and usually faster
12than dispatching, so the normal path is local with
13`pip install torch numpy biopython ProDy ml_collections dm-tree`; a GPU helps
14for batched campaigns.
15 
16## Running it
17 
18```bash
19pip install torch numpy biopython ProDy ml_collections dm-tree
20git clone --depth 1 https://github.com/dauparas/LigandMPNN.git ligandmpnn
21cd ligandmpnn
22sed -i 's/np\.int\b/np.int64/g' openfold/np/residue_constants.py   # repo pins numpy 1.23; alias removed in >=1.24
23bash get_model_params.sh ./model_params
24python run.py \
25  --model_type ligand_mpnn \
26  --checkpoint_ligand_mpnn ./model_params/ligandmpnn_v_32_010_25.pt \
27  --pdb_path complex.pdb \
28  --out_folder out \
29  --batch_size 8 --number_of_batches 4 \
30  --temperature 0.1 \
31  --fixed_residues "A45 A46 A47 A48"
32```
33…