proteinmpnn

1# ProteinMPNN
2 
3ProteinMPNN is the default inverse-folding step in the binder pipeline: a
4message-passing network that sees backbone geometry only, so it is the right
5choice when the design surface is protein–protein and the wrong one as soon as
6a ligand, nucleic acid, or metal is part of the interface — `ligandmpnn` adds
7those atoms to the graph with a near-identical CLI, and `solublempnn` swaps in
8weights trained on soluble structures for an expression-biased prior. Code and
9weights are MIT (github.com/dauparas/ProteinMPNN). The model is small enough
10to run on CPU — for a handful of sequences on one backbone that is seconds and
11usually faster than dispatching a remote job; a GPU helps for batched
12campaigns (hundreds of backbones or large `--num_seq_per_target`). Either way
13the repo is cloned in-job — there is no PyPI dist and the checkpoints are
14bundled in the repo.
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 --sampling_temp "0.1"
25```
26 
27Two flags trip almost everyone the first time. `--sampling_temp` is parsed as a
28space-separated string so one run can sweep several temperatures; a single
29value needs no quoting, but a multi-value sweep must be quoted
30…