ProteinMPNN
Design amino-acid sequences for a protein backbone structure. Finding a sequence from a structure is called inverse folding.
API/CLI tool ID: proteinmpnn
Understanding inputs and outputs
The backbone is the connected set of atoms forming a protein chain’s common framework. ProteinMPNN takes a structure file containing that backbone and proposes amino acid sequences that may fit it. Its input-to-output direction is the reverse of tools that predict structures from sequences.
Use designedResidues to specify positions to redesign for each chain. An empty object requests full design; to change only selected regions, specify chain IDs and residue numbers from the input structure. numSequences sets the number of candidates, while samplingTemp controls diversity when sampling candidate sequences. It is not an experimental temperature.
score evaluates the model’s fit for the designed positions, while global_score covers all positions. A lower score can help compare candidates but does not guarantee stability or function. seq_recovery measures how closely designed positions match the input sequence, so selecting only high values may not suit a search for new sequences.
Input settings
| Setting | Format | Required | Default | Description and constraints |
|---|---|---|---|---|
pdbFile | pdb | Required | — | Backbone structure to design — Extensions pdb · cif |
designedResidues | selectMultichainResidues | Required | {} | Residues to design per chain (blank: design all) — Example {"B":"26 27 28 29 30 31 32 52 53 54 55 56 95 96 97 98 99 100 101 102"} |
numSequences | number | Optional | 8 | Number of sequences to generate — Minimum 1, Maximum 100 |
samplingTemp | number | Optional | 0.1 | Greater than 0, up to 1. Lower values concentrate on sequences the model prefers, but results won't always be identical. — Maximum 1 |
omitAA | string | Optional | "X" | Uppercase amino acids to exclude. Must include X, with no duplicates or spaces (e.g., CX). At least one standard amino acid must remain. — Maximum length 20, Example "CX" |
Output metrics
| Column | Display name | Range | Description |
|---|---|---|---|
score | MPNN score | — | Mean negative log-likelihood at designed positions. Lower means the model considers the sequence a better fit for this backbone. It isn't binding, stability or the probability of experimental success. |
global_score | Global score | — | The same score, computed over all positions |
seq_recovery | Sequence recovery | 0~1 | Fraction of designed positions that keep the input amino acid (0–1). Higher isn't necessarily a better design. |
See Reading results for metric definitions and interpretation. A range shows possible values; whether higher or lower is better depends on the metric.
Execution availability
Open an execution plan in your project to see available tool versions and costs. Review and approve the inputs, settings, and terms before running.
License
Code, model weights, and accompanying data have their own terms. The table summarizes terms from official sources; check your execution plan for the configuration you will use. License texts are available in Third-party notices.
| Component | License | Commercial use | Source | Notes |
|---|---|---|---|---|
| Code | MIT | Allowed, subject to notice requirements | License text | — |
| Model weights | MIT | Allowed, subject to notice requirements | License text | The weights are included in the same repository |
Checked on 2026-11-01.