Computational antibody design

Predict antibody binding. Rank leads by developability. Skip the phage display bottleneck.

Genolux maps CDR–antigen contact interfaces from sequence, scores each variant for affinity and manufacturability, and delivers a shortlist of therapeutic candidates in weeks — not months of experimental screening.

~$800K

median cost of a single phage display campaign

Screening 109 variants to find 50 hits is the industry default.

The math only works because there was no alternative. Phage display runs 6–8 weeks, consumes $600K–$1.2M fully loaded, and produces a hit list you then have to characterize all over again for developability. We built Genolux because there is a better way now. Computational interface prediction maps the binding landscape from sequence — before a single phage is panned.

The platform

Three steps from sequence to shortlist

Submit your antibody sequence and target antigen. We run the full pipeline. You receive a ranked list of CDR variants scored for binding affinity (ΔΔG) and 7 physicochemical developability flags — before you send anything to synthesis.

Step 01 —

Interface Prediction

Input antibody FASTA + target antigen structure (PDB or homology model). Genolux runs AlphaFold2 Fv prediction and RosettaDock to identify CDR contact residues and buried surface area at the paratope–epitope interface.

Runtime: ~4 min/sequence

Step 02 —

CDR Optimization

Enumerate CDR loop sequence variants. Score each by predicted binding affinity (ΔΔG) using Rosetta energy functions trained on SAbDab + SKEMPI2 (30,000+ affinity measurements). Output: per-variant Kd prediction MAE < 0.4 kcal/mol.

Kd MAE < 0.4 kcal/mol

Step 03 —

Developability Ranking

Score every candidate against 7 physicochemical developability flags: aggregation propensity, charge patches, deamidation risk, oxidation sites, viscosity index, half-life prediction, and expression yield. Pareto-optimal shortlist delivered as CSV + JSON.

7 developability flags

Methodology

Built on structure prediction + physics-based scoring

No black box — every score traces to an energy term or database observation.

Structural inputs

AlphaFold2-derived structural context

Raw antibody sequences are processed through AlphaFold2 Fv prediction to generate the variable domain structure. The predicted Fv is then docked against the target antigen using RosettaDock protocol, identifying contact residues and buried surface area.

Read methodology
Scoring model

Rosetta energy function scoring

CDR variant scoring uses Rosetta REF2015 energy function. The ΔΔG model is trained on SAbDab structural data and SKEMPI2 binding affinity measurements — 30,000+ experimentally measured interaction energies across protein complexes, with antibody-specific weighting.

Read methodology
Optimization

Multi-objective Pareto optimization

Affinity and developability are competing objectives — a tighter binder may aggregate at high concentration. We compute the Pareto front across affinity × developability × expression yield to deliver candidates that are optimal in the practical sense, not just highest-affinity.

Read methodology

Why it matters

Genolux vs. phage display

Specific numbers, by metric. This is the case for pre-screening computationally before committing to phage selection.

Metric Genolux (computational) Phage display (experimental)
Timeline 2–4 weeks to ranked shortlist 6–8 weeks per campaign
Variants screened Up to 500 CDR point mutations per run 109–1010 phage library
Cost per campaign $490–$5,000 (platform pricing) $600K–$1.2M (fully loaded)
Output format Ranked CSV + annotated PDB + JSON API Hit pool requiring follow-on characterization
Iteration speed New design round in hours Weeks per design-test cycle
Developability data 7 flags included in every run Requires separate biophysical assays

Workflows

Three antibody engineering workflows

Where CDR-antigen interface prediction maps to concrete decisions in an antibody discovery program.

01

Hit identification

Start from a target antigen sequence or structure and an antibody scaffold. Genolux enumerates CDR loop candidates, scores binding contacts, and returns a shortlist of 20 high-confidence CDR sequence variants in two weeks — before committing to library synthesis.

Narrow from antigen to 20 CDR candidates in 2 weeks

02

Affinity maturation

You have a hit with the right specificity but not enough potency. Submit the hit sequence and target up to 500 single-point CDR mutations for ΔΔG scanning. Genolux ranks every variant by predicted affinity improvement, surfacing the highest-impact positions to mutate first.

Rank 500 CDR mutations by predicted ΔΔG

03

Developability filter

Before sending a panel to wet-lab synthesis, run it through the 7-flag developability screen. Genolux flags aggregation-prone hydrophobic patches, deamidation-susceptible NG motifs in CDRs, charge clusters affecting viscosity, and sequences with low predicted CHO expression yield.

Flag high-risk sequences before synthesis

Early access

From early partners

We used Genolux to pre-screen 300 CDR H3 variants against a GPCR antigen target before running any phage selection. The top 15 from the computational rank all confirmed as functional binders in SPR — and 4 of them had developability scores that would have flagged them out of a standard phage campaign anyway.

Dr. Karim Adeyemi

Director of Antibody Engineering
Clinical-stage biologics company

The developability scoring alone is worth the subscription. We run every candidate through Genolux before we decide which sequences to send to our CMO. The aggregation and deamidation flags have caught three sequences that would have failed late-stage biophysical characterization. The specificity of the output — per-residue energy decomposition — is what makes it actionable for our scientists.

Dr. Sophie Laroche

Principal Scientist, CMC
Early-stage biotech

Design better antibodies. Start with the sequence.

Early access is open to biologics discovery teams. Explorer tier is free — 10 analyses per month, no card required.