Absci’s Generative AI Approach Opens a Promising New Path for De Novo Antibody Design

In what the drug and target discovery company describes as “a major milestone for the biotechnology industry,” Absci has introduced a method for efficiently creating and validating de novo antibodies in silico using zero-shot generative AI.

Antibody drugs are widely used for treating infectious diseases, cancer, autoimmune diseases and inflammation. Existing de novo antibody design approaches however are time- and resource-consuming and usually have sub-optimal binding issues and poor developability attributes.

The Absci Corporation researchers address these issues in the new paper Unlocking de Novo Antibody Design With Generative Artificial Intelligence, which leverages the power of generative artificial intelligence for de novo antibody design in a zero-shot and controllable manner, dramatically reducing the time and resources required for the task.

The researchers’ main contributions can be summarized as follows:

• Designing antibodies against three different targets in a zero-shot fashion using generative deep learning models.
• Screening 400,000 antibody variants and characterizing 421 binders using surface plasmon resonance (SPR), discovering three that bind more tightly than the therapeutic monoclonal antibody trastuzumab.
• Generating sequences with high diversity and low similarity to known antibody sequences and variable structural morphology.
• Developing a highly controllable design approach for creating proteins with optimized exploitability and immunogenicity profiles that reduce downstream developability risks.

The proposed approach integrates novel generative modelling methods with high-throughput experimentation capabilities in a wet lab to validate de novo antibody design in a zero-shot setting. The team first generates new zero-shot HCDR3 sequences for known antibodies using generative AI models, demonstrating this capability using trastuzumab and its target antigen, the human epidermal growth factor receptor 2 (HER2). They then design some 440,000 unique HCDR3 variants of trastuzumab, screen them for binding to HER2, and functionally validate 421 binders using SPR.

The team’s zero-shot designs display tighter binding than trastuzumab, and the antibodies can be generated without additional affinity maturation, significantly reducing development time. The AI model designs also show high sequence diversity in amino acid length and identity.

Finally, the team demonstrates their approach’s ability to generate sequences with high affinities and naturalness scores in a zero-shot manner, validating its potential for bypassing parts of the traditional lead optimization process to save time and resources in the drug discovery process.

The researchers believe their breakthrough in in-silico antibody design could revolutionize the development of effective therapeutics for patients, with its controllability enabling the creation of customized molecules for specific disease targets that lead to safer and more efficacious treatments.

The code and HER2 binders are available on the project’s GitHub. The paper Unlocking de Novo Antibody Design With Generative Artificial Intelligence is on bioRxiv.

---

Author: Hecate He | Editor: Michael Sarazen

---

We know you don’t want to miss any news or research breakthroughs. Subscribe to our popular newsletter Synced Global AI Weekly to get weekly AI updates.