Enzyme campaigns
Build a better variant library before your next screen.
An enzyme campaign turns sequence and assay constraints into ranked candidates, constrained libraries, risk flags, and handoff-ready outputs for the next build-test round — for programmes with a lead enzyme, a defined objective, and a real path to test.
The design pack
A focused deliverable your wet lab can use immediately.
The offer stays intentionally narrow: one lead enzyme, one concrete optimization objective, and a fixed-scope campaign that results in a clear next plate.
Ranked mutation candidates
Prioritized with multiple scoring signals, your assay constraints, and explicit do-not-mutate boundaries.
Core library
A compact set of conservative, diverse candidates for teams working within a tight screening budget.
Expanded library
Broader coverage for teams that want more exploration while keeping the shortlist interpretable.
Risk flag report
Stability, expression, and design-space risks surfaced before the next build-test round.
Handoff files
CSV, FASTA, and structured outputs formatted for synthesis vendors or internal wet-lab workflows.
Round-two learning loop
Assay results fold back into the next prioritization, so each round becomes more informed.
Campaign shape
From design space to a testable library.
Candidates are generated at designable positions, scored by an ensemble of independent signals, filtered against structure, and finally composed into libraries sized to your screening budget.
Screening funnel
- Generated1,240
Substitutions at designable positions, with active-site and disulfide positions locked out.
- Screened260
Multi-signal scoring: language-model log-odds, stability heuristics, pH and aggregation terms.
- Structurally evaluated140
Solvent accessibility and burial context from structure, with low-confidence regions filtered.
- Recommended96
A core plate plus an expanded set, sized to your screening budget with rationale per pick.
Scoring signals
Language-model score
Masked-marginal log-odds from a protein language model as a fast, general sequence signal.
Stability heuristics
Structure-aware rules on burial, volume matching, and secondary-structure context.
pH and surface behaviour
Surface charge treatment when the objective involves working outside neutral pH.
Aggregation risk
Hydrophobic-patch penalties from sliding-window hydropathy.
Signals are normalized and weighted rather than collapsed early, and candidates where independent signals disagree are flagged rather than silently ranked.
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
Two-stage design engine
A constrained first round, then a smarter second.
The initial campaign covers Stage 1. Stage 2 begins once assay measurements are available.
Stage 1
Initial candidate generation and library design
- Generate and rank candidates under assay and library constraints.
- Combine objective rules, allowed or forbidden positions, and multiple scoring signals.
- Deliver a compact first-round library with rationale and risk flags.
Stage 2
Assay-informed round-two optimization
- Ingest first-round assay measurements and observed hits or misses.
- Update prioritization using project-specific signals and revised hypotheses.
- Propose a more informed round-two shortlist.
Fit boundaries
Better qualification makes the work more useful.
Good fit
- You already have a lead enzyme or a starting variant family.
- You know the property you want to improve and how you will measure it.
- You can test shortlisted variants experimentally after the campaign.
- You want better prioritization before committing the next plate.
Not a fit
- Open-ended discovery without a starting enzyme.
- Guaranteed-outcome expectations for a single round.
- Fully outsourced large-scale campaigns from day one.
- Buyers looking for self-serve software instead of scoped technical work.
FAQ
The questions technical buyers ask first.
Do you need prior assay data?
No. A lead enzyme, a concrete property objective, and a realistic test plan are enough to scope an initial campaign.
Do you need a structure?
Not always. A campaign can start from sequence plus assay context and incorporate structural information when it is available and useful.
What do you need from us to start?
The lead enzyme or variant family, the property to improve, the assay context, the screening budget, and any hard constraints is enough for an initial fit review.
Do you guarantee improved variants in one round?
No. The work is designed to improve prioritization quality and reduce low-information picks, not to promise a single-round outcome.
How do you handle confidentiality and sequence data?
Campaigns can start under NDA. A lightweight fit conversation can happen before sensitive technical details are shared, and the first email does not need to include proprietary sequence information.
How is this validated?
Ranking quality is being benchmarked on public held-out data — see the benchmark status on this page. Results will be published when they exist, including negative ones.
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.