A Glimpse into De Novo Antibody Design

At Galux, we are redefining how drugs are created. Instead of relying solely on traditional drug discovery where years of trial-and-error screening are required and many challenging targets remain out of reach, we are pioneering an AI-driven approach to design proteins from the start.

Our proprietary GaluxDesign platform integrates deep physical chemistry knowledge with cutting-edge AI, enabling the precise, rational designing of therapeutic proteins tailored to specific targets. This capability holds particular promise for the earliest stages of drug discovery and development, where designing the right molecule can set the trajectory for the entire process.

Today, we’re taking you inside one such example: an antibody we designed de novo against PD-L1, and sharing new experimental results that validate both its novelty and the precision of our AI design process.

1. Looking back at our March results

This March, we announced a major milestone:
De novo antibody design for six therapeutic targets (PD-L1, HER2, EGFR (S468R), ACVR2A/B, ALK7, FZD7, a first in the field to successfully demonstrate AI platform’s generalizability across diverse targets. [Read to full study]

Some key highlights include:

Successfully designed antibodies against a target lacking an experimentally resolved
structure.
Designed antibodies demonstrated biophysical, functional, and developability profiles comparable to those of a commercial therapeutic antibody.
Designed antibodies distinguished closely related target protein subtypes and mutants differing by just a single amino acid, highlighting the platform’s atomic-level precision.

The study drew meaningful attention across the industry, as many continue to explore how AI can contribute to therapeutic discovery.

Among them was a PD-L1-targeting antibody, GX-aPDL1-3, designed by GaluxDesign. It demonstrated binding affinity and thermal stability comparable to that of atezolizumab.

2. Structure determined

We’ve now experimentally determined the complex structure of PD-L1 and GX-aPDL1-3 using Cryo-EM.

This allows us to take a closer look at two key aspects: the novelty of the designed antibody, and the accuracy of the designed structure.

3. Novelty in sequence and binding mode

From a sequence perspective, GX-aPDL1-3 is highly distinct from any PD-L1 antibody in the Protein Data Bank (PDB) showing just:

43% sequence identity in the CDR loops
27% sequence identity in the critical H3 loop

Structurally, GX-aPDL1-3 binds to the target in a binding mode distinct from any PD-L1 antibody in PDB, demonstrating GaluxDesign’s ability to generate not just new sequences, but entirely new ways of interacting with a target.

4. Precision of the designed structure

Every antibody generated by GaluxDesign is modeled with a predicted binding pose during the design process.

We compared the designed structure of GX-aPDL1-3 with the experimental structure. The interface RMSD between the two structures was only 1.1 Å, demonstrating that the designed structure matches the experimental structure in atomic-level precision.

5. Looking ahead

Although this is only a single example, it offers a glimpse into how AI-driven protein design could reshape protein therapeutic development. In future, the target-antibody complex structure could be the data that is naturally derived during the design phase.

At Galux, we’re building a world where drugs are no longer discovered, but they’re precisely and rationally designed.

This is just the beginning. Follow along as we continue to explore the frontiers of AI-driven therapeutic design. More results are on the way, so stay tuned.

If you share our vision for reimagining drug discovery and development through AI, or simply want to explore what’s possible together, we’d love to hear from you.

Every conversation could be the spark for the next breakthrough.

Newsroom