Scoped campaigns · Antibodies and enzymes
Design better protein variants before your next experiment.
Foldry turns sequence and structure plus your real experimental constraints into computational design and prioritization — and hands back a ranked, lab-ready candidate set for the next round.
A lead molecule
An antibody, an enzyme, or a variant family already worth optimizing.
A measurable objective
Affinity, stability, expression, developability — something you can actually assay.
A path to test
A campaign supports a real build-test round. It does not replace one.
Common friction
- Plate budget disappears into near-duplicate variants
- Intuition-led picks dominate the shortlist
- Improving one property quietly breaks another
- The team cannot explain why each candidate made the plate
- Assay results do not feed back cleanly into the next round
Where we work
Two molecule classes, one method.
The workflow is the same shape in both: a constrained design space, a screening cascade, and a defensible set at the end. The models and the property checks differ by molecule.
Flagship
Antibody lead optimization
A lead antibody and its antigen context become a ranked set of 24, 48, or 96 variants for the next wet-lab round — screened for binding, developability, and humanness together rather than one at a time.
- CDR-focused variant generation under locked positions
- Interface analysis against the antigen
- Developability and humanness filters
- Pareto and diversity selection for the plate
Established offering
Enzyme optimization
A lead enzyme and a measurable property objective become a constrained, testable library — ranked candidates, explicit risk flags, and handoff files your wet lab can use without reformatting.
- Structure-informed designable-position selection
- Multi-signal scoring ensemble
- Core and expanded library design
- Assay-informed second round
How a campaign works
Scoped work with a defined end state.
A campaign is a fixed piece of technical work, not an open-ended engagement or a software subscription. It ends in a candidate set and a decision.
01
Scope
One molecule, one concrete objective, and the constraints that are real: locked positions, assay format, screening budget, timeline.
02
Design
Generate a large candidate space, then narrow it with sequence filters, structural evaluation, and property models appropriate to the molecule class.
03
Prioritize
Rank across competing objectives instead of a single score, and select a set that is both strong and diverse enough to be informative.
04
Hand off
A candidate set sized to your plate, with per-candidate rationale, risk flags, and files formatted for synthesis or internal workflows.
05
Learn
Assay results from the round feed back into the next prioritization, so each cycle starts better informed than the last.
Example campaign
What went in, what ran, what came out.
This is the shape of an antibody lead optimization campaign: a large generated space narrowed by successively more expensive checks, ending in a set small enough to actually run.
Screening funnel
- Generated1,800
CDR-focused variants from inverse folding and language-model proposals, under your locked positions.
- Screened300
Fast sequence-level filters: liability motifs, charge and pI bounds, germline distance.
- Structurally evaluated96
Complex modelling and interface analysis against the antigen, plus antibody-specific geometry checks.
- Recommended24
Pareto-optimal across binding, developability and humanness, then spread for diversity.
Candidate landscape
Selection is made across competing objectives rather than on a single score, then spread for diversity so a failed hypothesis still leaves the round informative.
Candidate-level output
| Variant | Region | Signals | Decision |
|---|---|---|---|
| HC S103T / Y105F | CDR-H3 | Interface contact gain · no new liability motif | Core |
| HC T57A | CDR-H2 | Predicted contact retained · germline-proximal | Core |
| LC N92Q | CDR-L3 | Removes deamidation motif · neutral on predicted binding | Core |
| HC G55E / LC S31R | CDR-H2 + CDR-L1 | Larger predicted gain · charge shift near interface | Expanded |
| HC W47L | Framework | Aggregation-risk flag · framework position | Held |
Benchmarks in progress
Foldry does not claim validated wet-lab outcomes it does not have. These are the public benchmarks currently being run and the metrics that will be published — including negative results.
AbBiBench
In progressAntibody binding-variant ranking on held-out labels
Enrichment · precision@k · Spearman, against single-model and naive baselines
FLIP2 alpha-amylase
In progressHeld-out library selection on an official train/test split
Enrichment at fixed budget · nDCG@k · hits-vs-budget against random libraries
Sample campaign report
A full example report — objective and constraints, methods, ranked candidates, library composition, risk flags, and handoff instructions — is available on request.
Technical approach
An evolving stack of the best available models.
Foldry does not claim a proprietary foundation model. The work is in composing the right models for a given problem, constraining them with real experimental limits, and validating the ranking that comes out. The value is the workflow and the experimental decision, not access to any single model.
Structure prediction
Antibody and general protein structure, complex modelling where the interface matters.
Inverse folding and sequence design
Structure-conditioned proposals at the positions you allow.
Variant generation
Language-model and rule-driven candidate spaces under hard constraints.
Binding and interface analysis
Contact and geometry assessment against the target.
Developability
Aggregation propensity, charge and pI, expression and liability motifs.
Humanness
Germline distance and humanness scoring for antibody campaigns.
Multi-objective ranking
Trade-offs made explicit rather than collapsed into one number.
Experimental feedback
Measured outcomes folded into the next round's prioritization.
Technical details
Campaigns draw on open and published tooling — antibody numbering and region assignment, structure and complex prediction, inverse-folding and protein language models for proposal generation, and established developability and humanness metrics. Model selection is decided per campaign against the objective and the constraints, and the specific stack used is documented in the campaign report rather than treated as a black box.
Engagement
Fixed-scope first campaign.
Pilot campaigns are scoped against the objective and the experimental design, so the price reflects the actual problem rather than a pre-set package.
Discovery review
Free
Fit check and scope review
- Molecule and objective fit assessment
- Constraint and assay review
- Feasibility and timeline view
Pilot campaign
Scoped to the problem
Priced against the objective and the experimental design
- Candidate generation and multi-objective ranking
- Core and expanded candidate sets
- Risk and liability review
- Wet-lab handoff files
Ongoing campaigns
Custom
For multi-round programmes after a first campaign
- Round-over-round prioritization
- Assay feedback integration
- NDA and custom scope support
Start the conversation
Bring a real molecule and a real next experiment.
The first step is a scoped conversation about whether a campaign would change what you put on the next plate.
Useful to include
- 01The lead molecule or variant family you are working from.
- 02The property you want to improve and how you measure it.
- 03Screening budget, timeline, and any hard constraints.
NDA available before sharing sequences or other sensitive detail.