GenovaX — AI-Designed Nanobody Binders Against a Benchmark Target (Methodology Demonstration)

Ranked shortlist (10 designs, by binding confidence)

Binding confidence spans ~0.37 to ~0, so the model discriminates rather than rubber-stamps. Design 8 has the highest interface/fold confidence but near-zero binding confidence — which is why the shortlist ranks on the binding objective, not any single metric.

Lead candidate — R7gl

GenovaX · computational prediction

R7gl (lead)

OverviewConfidenceInterface

Domain overview

Rendered view of the predicted target and nanobody complex with the structural regions identified in the underlying analysis.

Predicted metrics

Binding confidence

0.37 (top of 10)

Interface ipTM

0.87

Nanobody mean pLDDT

~85

Paratope in CDRs

22 / 25

Developability

Grade A (clean)

Predicted contact sets

Paratope residues

25

Epitope residues

28

Underlying model file

Download predicted CIF

Predicted binding is CDR-driven — 22 of 25 paratope residues fall in the CDR loops. Clean developability profile (no CDR liabilities; only the 2 canonical cysteines).

Second candidate — 33Tc

GenovaX · computational prediction

33Tc

OverviewConfidenceInterface

Domain overview

Rendered view of the predicted target and nanobody complex with the structural regions identified in the underlying analysis.

Predicted metrics

Binding confidence

0.25 (#2 of 10)

Interface ipTM

0.85

Nanobody mean pLDDT

~85

Paratope in CDRs

13 / 16

Developability

Grade C (moderate)

Predicted contact sets

Paratope residues

16

Epitope residues

20

Underlying model file

Download predicted CIF

Strong binder but developability filtering flags three CDR liabilities — including an N-glycosylation sequon inside a CDR, plus a deamidation site and an oxidation-prone methionine in a CDR — so it is a conditional #2 pending liability-removal engineering.

Developability filtering (why the ranking changed)

We screened all designs for standard antibody manufacturability liabilities — CDR N-glycosylation sequons, deamidation/isomerization motifs, oxidation-prone Met/Trp, extra cysteines, charge/pI, hydrophobicity (Therapeutic-Antibody-Profiler-style heuristics, not lab assays). The lead R7gl is best on BOTH binding and developability; the #2 by binding (33Tc) was demoted to conditional because of CDR liabilities. Revised recommendation: advance R7gl as sole lead; hold 33Tc pending engineering; keep So7D as a clean backup.

Honest limitations & next steps

• In-silico only; no expression, purification, or affinity has been measured.
• Absolute binding-confidence values are modest (top ~0.37); rank-order is the useful signal, not the magnitude.
• Lysozyme is a well-characterized structural-biology benchmark, not a Planckchron therapeutic target.
• Next step: wet-lab validation of the top 2–3 candidates (express → SPR / BLI / ELISA).
• Computational predictions require experimental validation. Not medical, clinical, or investment advice.