Vanilla full-backbone checkpoints
General-protein design with v_48_002, 010, 020, or 030.
Start this methodGeneration method
Run ProteinMPNN online
Design protein or antibody sequences for a backbone
Generate sequence candidates compatible with a full backbone using standard ProteinMPNN or the antibody-specific AbMPNN model, preserve chosen positions, and compare recovery, mutations, and model scores.
Choose an approach
Scientific approaches
Select the analysis that best matches your scientific question. Each approach opens with its relevant inputs and controls.
Alternative vanilla and soluble checkpoints
Four vanilla and four soluble-protein checkpoints spanning 0.02–0.30 Å training noise.
Start this methodAbMPNN antibody design
Heavy-chain or paired heavy/light redesign with explicit chain roles and fixed antibody positions.
Start this methodPrepare
Full-backbone protein or antibody structures
Start from a Structures Dataset containing PDB files with complete N, Cα, C, and O backbone coordinates. For AbMPNN, identify the heavy chain and optional paired light chain explicitly; every other chain remains fixed context.
- Complete full-backbone PDB coordinates
- Standard method: one or more designable protein chains
- AbMPNN: one heavy chain and an optional distinct light chain
- Up to 2 structures and 2,000 total residues per analysis
Design chains
["A"]Fixed positions
{"A":[23,45,102]}AbMPNN chain roles
heavy H · light LHow the analysis starts
Choose a Project, open a compatible Dataset, then select the rows you want to analyze. Ubi will open this method with the compatible controls and column mappings ready to review.
Configure the scientific method
These controls appear in the Dataset analysis panel, where values can be checked against the actual input before the run starts.
Method
Standard ProteinMPNN or AbMPNN
Use the general or soluble checkpoint families for proteins; use the distinct AbMPNN action for antibody variable-domain chains.
Standard checkpoint
Vanilla or soluble · 0.02, 0.10, 0.20, or 0.30 Å
For standard ProteinMPNN, choose the training scope and noise level; the exact checkpoint is retained with every candidate.
Design scope and chain roles
Selected protein chains or antibody H/L · fixed positions
Choose which protein chains can change, or declare AbMPNN heavy/light roles; preserve selected one-based sequence positions.
Candidates and sampling
1–16 sequences · temperature, seed, noise, omissions
Control candidate count, diversity, repeatability, backbone perturbation, and excluded amino acids.
Review results as scientific outputs
Results open with the figures, structures, sequences, and metrics needed to answer the scientific question. Downloadable files remain available for downstream analysis.
How to interpret the result
- ProteinMPNN score is a negative-log-probability measure; lower values indicate sequences the model considers more compatible with the backbone.
- Sequence recovery is similarity to the source sequence, not an independent quality score. Interpret it in light of the intended redesign depth.
- AbMPNN is an antibody-specific prediction model, not an experimental binding or developability assay. Compare candidates only after confirming the heavy/light roles and fixed-position constraints.
- Review chain-resolved mutations and fixed-position compliance before advancing a candidate to structure prediction or experimental work.
1
Ranked sequence designs
Compare model score, global score, and sequence recovery.
2
Chain-resolved sequences and roles
Review designed chains without losing fixed context; AbMPNN results retain explicit heavy/light roles.
3
Exact mutation list
Inspect residue changes using both sequence and PDB residue positions.
Continue your analysis
Use a complementary method on the same Project data.
Method scope and limitations
- Standard ProteinMPNN includes vanilla and soluble full-backbone models at four training-noise levels.
- AbMPNN is a distinct antibody-specific model, not a standard ProteinMPNN checkpoint.
- CA-only, PSSM, tied-position, residue-bias, and probability-matrix modes are not available in this analysis.
