SolubleMPNN Solubility-Optimized Design

Prerequisites

How to run

First time? See [Getting started](../../docs/getting-started.md) to set up Modal and biomodals.

Option 1: Modal (recommended)

SolubleMPNN is the soluble model type within the LigandMPNN wrapper:

cd biomodals
modal run modal_ligandmpnn.py \
  --input-pdb backbone.pdb \
  --params-str "--model_type soluble_mpnn --number_of_batches 16 --temperature 0.1"

GPU: A10G default | Timeout: 900s default

Option 2: Local installation

git clone https://github.com/dauparas/ProteinMPNN.git
cd ProteinMPNN

# The soluble weights are selected with --use_soluble_model, not a model name
python protein_mpnn_run.py \
  --pdb_path backbone.pdb \
  --out_folder output/ \
  --num_seq_per_target 16 \
  --sampling_temp "0.1" \
  --use_soluble_model

Key parameters

Model weights

--modelname sets the training-noise level (v48002 = 0.02 A, v48010 = 0.10 A, v48020 = 0.20 A), not a solubility tier. Solubility is a separate weight set chosen with --usesolublemodel, available for v48010 and v48_020. Higher noise gives more sequence diversity.

Output format

output/
├── seqs/backbone.fa
└── backbone_pdb/backbone_0001.pdb

Sample output

Successful run

$ python protein_mpnn_run.py --pdb_path backbone.pdb --use_soluble_model --num_seq_per_target 8
Loading soluble model weights (v_48_020)...
Designing sequences for backbone.pdb
Generated 8 sequences in 2.1 seconds

output/seqs/backbone.fa:
>backbone_0001, score=1.31, global_score=1.24, seq_recovery=0.78
MKTAYIAKQRQISFVKSHFSRQLE...
>backbone_0002, score=1.28, global_score=1.21, seq_recovery=0.81
MKTAYIAKQRQISFVKSQFSRQLD...

What good output looks like:

• Score: 1.0-2.0 (lower = more confident)
• Reduced hydrophobic patches compared to standard MPNN
• Improved charge distribution

Decision tree

Should I use SolubleMPNN?
│
├─ What expression system?
│  ├─ E. coli → SolubleMPNN ✓
│  ├─ Mammalian → ProteinMPNN (PTMs matter more)
│  └─ Yeast → Either
│
├─ History of expression problems?
│  ├─ Yes, aggregation → SolubleMPNN ✓
│  ├─ Yes, low yield → SolubleMPNN ✓
│  └─ No → ProteinMPNN is fine
│
├─ What's in the binding site?
│  ├─ Small molecule / ligand → Use LigandMPNN
│  └─ Nothing / protein only → SolubleMPNN ✓
│
└─ Optimizing for expression?
   └─ Add --use_soluble_model to ProteinMPNN

Typical performance

Expected improvement: +15-30% solubility score vs standard ProteinMPNN.

Verify

grep -c "^>" output/seqs/*.fa  # Should match backbone_count × num_seq_per_target

Troubleshooting

Still insoluble: Confirm --usesolublemodel is set; redesign more positions or add explicit hydrophobic-residue bias Low diversity: Increase temperature to 0.2 Poor folding: Use standard ProteinMPNN and optimize later

Error interpretation

Next: Structure prediction for validation → protein-qc for filtering.