Applications
Two problems, one engine.
Proteinvue supports both therapeutic protein drug discovery and industrial enzyme process development, through the same generative sequence search.
Therapeutic protein design
Antibody CDR optimization
Improve binding affinity or reduce immunogenicity in CDR loops without full re-screening from scratch. When you have a lead antibody that binds the target but needs affinity maturation or developability improvement, Proteinvue generates ranked CDR variants guided by learned sequence-function co-variation rather than random mutagenesis. Designed for small drug discovery teams running 96-well plate assays, not enterprise CROs with unlimited synthesis budget.
Nanobody and VHH engineering
The smaller scaffold of nanobodies makes manual iteration harder because every residue matters more. The VHH CDR3 loop is unusually long and adopts conformations not seen in conventional antibodies, which means co-variation patterns specific to camelid single-domain antibodies must be captured correctly. Proteinvue accounts for VHH-specific evolutionary signals in candidate generation, producing shortlists that respect the geometric constraints of the scaffold rather than treating it as a truncated conventional antibody.
Cytokine and scaffold engineering
Multi-function objectives, such as improving receptor binding on one face of a cytokine while reducing off-target signaling on another, require simultaneously satisfying constraints at multiple sequence positions. Hand-design struggles here because the epistatic couplings are non-obvious. Generative search samples from regions of sequence space where the model has observed co-evolved solutions to multi-site constraints.
Industrial enzyme engineering
Thermostability and activity co-optimization
Increasing thermal stability often reduces catalytic activity. The mutations that rigidify the protein near the active site can also constrain the conformational dynamics needed for substrate turnover. Proteinvue can identify sequence regions where the model has learned that thermostable variants have preserved or even enhanced activity, based on evolutionary signal from thermophilic relatives. This does not guarantee both properties simultaneously, and we are explicit about that, but it focuses experimental effort on the candidates most likely to thread the needle.
Substrate specificity broadening or narrowing
When a specialty chemical process needs an enzyme to accept a slightly different substrate, or to stop accepting a competing one, the design problem is narrower than therapeutic engineering but the stakes are equally high. A substrate specificity shift typically involves a small number of active site residues, which makes the co-variation signal particularly informative. Proteinvue generates candidates that modify specificity pocket geometry while preserving the catalytic mechanism.
Soluble expression improvement
Enzymes that express as inclusion bodies or aggregate in E. coli or yeast fermentation waste considerable process development time. Sequence-level determinants of solubility and expression are learnable from co-variation in well-expressed homologs. Proteinvue can generate candidates that preserve catalytic function while altering surface charge distribution and hydrophobic patch exposure in ways that correlate with improved expression in bacterial or yeast hosts.
What changes with Proteinvue
| Workflow step | Without Proteinvue | With Proteinvue |
|---|---|---|
| Sequence candidates to synthesize per cycle | 30 to 80 variants, prioritized by expert intuition and limited prior assay data | Top 5 to 10 ranked candidates, each with a predicted fitness score and rationale |
| How assay queue is populated | Full 96-well plates, many redundant variants or positions that a co-variation signal would have deprioritized | Targeted subset, informed by learned sequence co-variation, MSA depth, and your prior assay data |
| Time to first ranked shortlist | Days to weeks of expert design time per cycle, depending on team capacity | Hours from objective definition to ranked FASTA delivery via the platform |
| Coverage of fitness landscape | Biased toward regions near the reference sequence that human intuition can anticipate | Generative search reaches epistatic combinations that single-residue walks and intuition miss |
Which use case fits your program?
Elena answers initial questions directly. Tell us your target protein, your fitness objective, and what assay data you already have. We will tell you honestly whether training coverage is adequate and whether a retrospective validation makes sense.
Or write directly: [email protected]