MPNN Sequence Design

Design amino-acid sequences for fixed protein backbones with ProteinMPNN, LigandMPNN or SolubleMPNN. Each backbone is held fixed while sequences are sampled for it, optionally keeping chosen chains or residues, redesigning only some, avoiding or favouring amino acids, tying symmetric chains together and packing side chains onto the designs. LigandMPNN also conditions on ligand, nucleotide and metal atoms in the structure. Every design carries a confidence and its recovery of the input sequence; recovery shows how conservative a design is, not how good it is.
0 backbones · 0 designs
Estimated minimum cost: $—

Examples

Backbones

Upload up to 50 protein structures (PDB or mmCIF), or fetch one from the PDB by its ID. Every backbone is held fixed and gets its own set of sequences. Nucleic-acid chains and ligands stay in the file as context for LigandMPNN; only protein chains are designed.

or drop files here

Model and sampling

General protein backbones.

1–256. Each sequence is sampled independently for the backbone.

0.01–1. Lower values give more conservative sequences; higher values give more diverse ones.

Set it to reproduce a run; leave it empty for a random seed.

Constraints

Chains and residues use the numbering of the structure files and apply to every backbone. PDB uploads are checked here; mmCIF files are checked when the job runs.

For example A, B. Leave empty to design every protein chain; chains left out stay fixed as context.

One-letter codes the designs may not use, for example C M.

Residue:value pairs added to the log-probability, for example A:-1.0, P:2.5. Positive favours a residue, negative disfavours it (±20 at most).

Chains of a symmetric assembly receive the same sequence; every chain needs the same residue numbers.

Adds a structure with modelled side chains for every design; limited to 200 designs per job.

LigandMPNN only.