StaB-ddG
Predict mutation effects on protein binding
Input
StaB-ddG webserver overview
StaB-ddG predicts mutation-induced changes in protein-protein binding free energy from a wild-type complex structure. It uses the final stabddg.pt checkpoint and supports single or combined substitutions, with one mutant per run or multiple mutants and complexes in a CSV.
Results are binding ΔΔG values in kcal/mol. Negative values indicate stronger predicted binding; positive values indicate weaker predicted binding. These predictions describe binding effects, not protein folding stability.
Pricing
Runs cost 27 credits per minute of measured runtime. The minimum reservation is 27 credits, not a minimum final charge. Completed runs are charged in proportion to elapsed time, rounded up to a whole credit; unused reserved credits are returned. Set a spending limit before submission. Each batch job is metered separately.
Inputs
| Mode | Required inputs | Formats and limits |
|---|---|---|
| One mutant | Wild-type complex, Mutation(s), and Binding partners | One .pdb file or an RCSB PDB structure; 50 MiB per uploaded file. |
| Mutation CSV | Complex PDB files and Mutation CSV | One or more .pdb files plus an uploaded .csv or pasted CSV text; 50 MiB per uploaded file. |
Mutation notation combines the wild-type amino acid, chain ID, position, and replacement amino acid. For example, EA63Q means E to Q at position 63 of chain A. Positions are chain-relative and start at 1; structures are not automatically renumbered. Commas join substitutions in a single combined mutant.
Binding partners are two chain groups separated by an underscore: ABC_DE means chains A, B, and C bind chains D and E. Chain IDs are individual letters or digits.
In CSV mode, the required, case-sensitive columns are #Pdb and mutation:
#Pdb,mutation
1AO7_ABC_DE,"EA63Q,QD30V,KA66A"
1C1Y_A_B,KB11MThe #Pdb value combines the PDB filename without .pdb and the two chain groups. The first row therefore requires 1AO7.pdb; the second requires 1C1Y.pdb. Quotation marks keep comma-separated substitutions in one CSV field. Additional columns are accepted.
Each structure occupies one input, with Add molecule providing additional inputs. PDB filenames must be unique by basename, begin with a letter or digit, and contain only letters, digits, periods, hyphens, or underscores. Optional precomputed binding-partner files, such as 1AO7_ABC.pdb and 1AO7_DE.pdb, can accompany the complexes. Inference has a 55-minute time limit.
Settings
Mutant and interface
| Parameter | Type | Default | Description |
|---|---|---|---|
| Input mode | enum | single | single selects One mutant; csv selects Mutation CSV. |
| Mutation(s) | string | EA63Q,QD30V,KA66A | Single mode only; one substitution or a comma-separated combination. |
| Binding partners | string | ABC_DE | Single mode only; two chain groups separated by one underscore. |
Sampling and execution
| Parameter | Type | Default | Description |
|---|---|---|---|
| Monte Carlo samples | integer | 20 | At least 1; samples averaged within each trial. |
| Independent trials | integer | 1 | At least 1; produces separate pred_1 through pred_N columns. |
| Random seed | integer | 0 | Seed supplied to PyTorch inference. |
| Backbone noise (angstrom) | number | 0.1 | Backbone-noise setting in angstroms. |
| Token batch size | integer | 10000 | At least 1; residue tokens per batch, which must accommodate the complex length. |
| Inference device | enum | cuda | cuda uses a GPU when available and otherwise CPU; cpu selects CPU. |
| Prediction filename | string | output | Filename stem without directory separators; .csv is appended. |
| Job name | string | optional | Label used in job history. |
Outputs
The Binding ΔΔG (kcal/mol) table displays the predictions. The Files tab provides individual downloads, and Download all collects the files in a ZIP.
| Download | Contents |
|---|---|
native/output.csv | Prediction table with native headers, row order, numeric precision, and CSV index; the stem follows Prediction filename. |
native/input.csv | Prepared mutation table, including additional submitted CSV columns. |
PDB files under native/ or native/pdbs/ | Wild-type complex and binding-partner structures, including supplied partner files or extracted chains. |
stab_ddg.log | Inference command and execution log. |
provenance.json | Settings, source and model identity, checksums, units, and sign convention. |
Returned PDB files represent the wild-type complex and its binding partners. StaB-ddG does not generate mutant 3D structures.
Understanding results
| Field | Meaning |
|---|---|
Name | Complex and chain-group identifier, displayed as Complex interface. |
Mutation | Substitutions scored together in that row, displayed as Mutation(s). |
pred_1 through pred_N | Binding ΔΔG in kcal/mol for each independent trial, averaged over that trial's Monte Carlo samples. |
Trials remain separate and are not averaged together. Values near zero indicate a small predicted change relative to wild type. The prediction columns are energy estimates, not confidence scores.
Related tools
PRODIGY
Predict protein-protein binding affinity from structure
DeepRank-GNN-esm
Score every protein interface with ESM-2 and a graph neural network
IPSAE
Score interprotein interactions in AlphaFold and Boltz predictions
PLIP
Profile protein-ligand interactions from a PDB complex structure.
AllMetal3D
Predict metal and water binding sites in proteins with 3D CNN models.
DockQ
Assess docking quality between model and native structures
DSSP
Assign helices, sheets, turns, bends, and PPII structure from atomic coordinates.
LocScale
Physics-informed local sharpening for cryo-EM density maps.
MolProbity
Validate protein structures with clashscore, Ramachandran, rotamer, and geometry checks.
PDBsum
Generate structural summary reports and diagrams for a single protein PDB structure.