• Pioneering AI technology successfully designs antibodies de novo for six therapeutic targets, including one without an experimentally resolved structure

Galux Inc., a South Korean startup specializing in AI-driven protein therapeutics design, has published a study showcasing the capabilities of its AI platform, GaluxDesign, in de novo antibody design. This research marks an important milestone in AI-driven antibody discovery, demonstrating the successful design of antibodies against six distinct therapeutic targets, including a target without an experimentally resolved structure.

"De novo antibody design has been one of the major challenges in AI-based drug discovery, requiring atomic-level precision to ensure specific binding to target epitopes," said Chaok Seok, CEO of Galux. "A few previously reported cases of de novo antibody design have shown only limited success, particularly in terms of target diversity and binding affinity. Our findings prove that AI can reliably generate novel antibodies with high precision, specificity, and sensitivity across diverse therapeutic targets."

A graphical representation verifying the binding strength between the antibodies designed by GaluxDesign and each therapeutic target.

The study presents an AI-driven approach to de novo antibody design. The team identified binders from a yeast display single-chain variable fragment (scFv) library comprising approximately one million designed antibody sequences, followed by screening binders against a target protein. Binders with varying binding strengths were identified for six targets, including PD-L1, HER2, EGFR(S468R), ACVR2A/B, FZD7, and ALK7.

Notably, the team successfully designed antibodies targeting ALK7, a protein without an experimentally resolved structure, as well as antibodies targeting a novel epitope of FZD7 discriminating subtypes FZD1 and FZD5. This highlights the platform's broad applicability for novel antibody discovery.

The designed PD-L1 targeting antibody showed outstanding binding affinity (KD=9.0pM) and developability comparable to the commercial therapeutic antibody Atezolizumab. Additionally, antibodies designed for EGFR(S468R) displayed exceptional specificity, precisely distinguishing the mutant from wild-type EGFR based on a single amino acid difference, which is an essential feature for minimizing off-target effects in early-stage drug development.

"This study underscores the potential of AI-driven antibody design and its broad applicability in antibody discovery," added Seok. "By continuously advancing our AI platform with a deep understanding of atomic-level protein interactions, we aim to transform therapeutic drug discovery, improving both efficiency and success rates."