Binding ΔΔG

Rank antibody variants on an antibody-antigen complex with a zero-shot ΔΔG proxy. Every variant is scored against the fixed parent with ProteinMPNN inverse-folding log-probabilities on the same backbone — no mutant structure and no per-target training — and each row carries a mean and a standard deviation across stochastic passes in kcal/mol, lowest first. The score is computed on the bound complex only, so it mixes folding and binding effects. Predictions support prioritization within one engineering campaign; they are not absolute affinity measurements.
0 variants ready
Estimated minimum cost: $—

Examples

Complex and parent

Upload the antibody-antigen complex as a PDB file. The parent sequence must match one chain of that complex exactly — tags included — because the model scores the variants against the chain it is matched to.

or drop a file here

Parent chain sequence

One-letter codes or one FASTA record, up to 5 000 residues. Paste the sequence of the chain you are mutating, exactly as it appears in the complex.

or drop a file here

Variants

List the variants to rank against the parent. Each variant is scored on the same complex backbone, so only the mutated residues move. At most 50 variants per run.

One variant per line, written as the residue you expect, its position, and the replacement — T315I — joined with + to combine several in one variant. Positions are the complex chain's residue numbers; each expected residue is checked against that chain when the job runs, so a numbering offset fails with the residue actually found.

No variants yet.

Numbering and sampling

First residue number of the parent sequence. Literature numbering often differs from a construct's indices; a wrong value fails the residue check instead of mutating a different site.

Scoring passes per variant (1–32). Each pass draws fresh decoder noise; the table reports the mean and standard deviation of the predicted change across passes.