Chai-2 AI Revolutionizes Antibody Design with 16% Success Rate in Zero-Shot Discovery

Breakthrough in Antibody Design with Chai-2

The Chai Discovery Team has launched Chai-2, a cutting-edge multimodal AI model designed for zero-shot de novo antibody design. This innovative platform achieves a remarkable 16% hit rate across 52 novel targets, using no more than 20 candidates per target. This outperforms previous methods by over 100 times and delivers validated binders in less than two weeks, eliminating the need for extensive high-throughput screening.

How Chai-2 Works

Chai-2 integrates an all-atom generative design module alongside an advanced folding model that predicts antibody-antigen complex structures with twice the accuracy of its predecessor, Chai-1. It operates without requiring any target-specific tuning and can generate sequences for various antibody formats such as scFvs and VHHs. The model can also incorporate epitope-level constraints and supports cross-reactivity design between species like human and cynomolgus monkeys.

Impressive Experimental Results

Tested on 52 targets with no prior known binders in the Protein Data Bank, Chai-2 achieved:

• A 15.5% average hit rate across all antibody formats
• 20% hit rate for VHHs
• 13.7% hit rate for scFvs
• Validated binders for 26 targets, including historically challenging ones like TNFα

Many binders demonstrated high affinity with picomolar to low-nanomolar dissociation constants, signaling strong and specific interactions.

Unique Features and Design Diversity

Chai-2 produces antibody designs that are structurally and sequentially novel. None of the generated designs had less than 2Å RMSD compared to known structures, and all complementarity-determining regions (CDRs) showed more than 10 edits difference from known antibodies. The binders clustered into multiple structural groups per target, indicating conformational diversity. Off-target binding was low, and polyreactivity profiles were comparable to approved clinical antibodies like Trastuzumab.

Customizable and Flexible Antibody Engineering

This platform allows for targeting multiple epitopes on a single protein, generating different antibody formats, and designing cross-species reactive antibodies within a single prompt. A case study demonstrated Chai-2’s ability to design an antibody binding nanomolar affinity to both human and cyno protein variants, which is crucial for preclinical and therapeutic applications.

Impact on Drug Discovery

Chai-2 drastically shortens the biologics discovery timeline from months to weeks by producing experimentally validated leads in one design cycle. Its success rate, novelty, and flexibility signify a paradigm shift in molecular engineering, opening doors for computational-first design approaches. Future developments may include bispecific antibodies, antibody-drug conjugates (ADCs), and optimization of biophysical properties like viscosity and aggregation.

Chai-2 sets a new standard for generative AI models in real-world drug discovery, demonstrating the potential to revolutionize therapeutic development beyond antibodies to miniproteins, enzymes, and small molecules.