solublempnn

1# SolubleMPNN
2 
3SolubleMPNN is not a separate package — it is the ProteinMPNN architecture
4retrained on a soluble-PDB subset, which shifts the output distribution away
5from the surface hydrophobics that the full-PDB model happily places (because
6many of them are buried at crystallographic or membrane interfaces in the
7training set). Reach for it when the goal is soluble yield in a heterologous
8host; stick with `proteinmpnn` when native-like recovery matters more, since
9the soluble prior trades a few points of recovery for the surface bias. Code
10and weights are MIT (github.com/dauparas/ProteinMPNN, `soluble_model_weights`;
11also exposed via github.com/dauparas/LigandMPNN). The model is small enough to
12run on CPU — for a handful of sequences on one backbone that is seconds and
13usually faster than dispatching; a GPU helps for batched campaigns. Either way
14the repo is cloned in-job (no PyPI dist; checkpoints bundled).
15 
16## Running it
17 
18```bash
19pip install torch numpy   # if not already present
20git clone --depth 1 https://github.com/dauparas/ProteinMPNN.git proteinmpnn
21cd proteinmpnn
22python protein_mpnn_run.py \
23  --pdb_path backbone.pdb --pdb_path_chains "A" \
24  --out_folder out --num_seq_per_target 16 \
25  --sampling_temp "0.1" --use_soluble_model
26```
27 
28The runner uses repo-relative imports, so the `cd` line is load-bearing —
29invoking the script by absolute path from elsewhere fails with
30`ModuleNotFoundError`. If you want threaded designed-sequence PDBs as well,
31…