HanchorBio and National Yang Ming Chiao Tung University (NYCU) have launched a collaboration that merges generative AI with experimental validation to accelerate antibody discovery for immunotherapies. Rather than relying solely on computational predictions or broad library screening, the partnership creates a closed-loop system where AI designs antibody sequences, lab experiments test them, and the results feed back into the AI to improve future designs.

Why Does Combining AI Design With Lab Testing Matter?

Traditional antibody discovery workflows cast a wide net. Researchers build large libraries of candidate molecules and screen thousands of them to find promising leads. This approach works, but it's time-consuming and resource-intensive. The HanchorBio-NYCU collaboration takes a different path by using AI-driven models to predict and generate antibody sequences against selected disease targets before any lab work begins.

The key advantage is efficiency. By narrowing the candidate space earlier in the process, researchers can focus experimental validation efforts on the most promising options rather than testing hundreds of mediocre candidates. This approach may reduce the burden of early screening and strengthen the overall quality of biologics entering development pipelines.

"This collaboration reflects our broader strategy to build differentiated discovery capabilities that can improve the quality, speed, and scalability of biologics innovation. By combining AI-enabled design with experimental validation, we aim to create a more efficient and adaptive discovery framework to support next-generation biologics," said Wenwu Zhai, Chief Science Officer of HanchorBio.

Wenwu Zhai, Chief Science Officer, HanchorBio

How Does the Closed-Loop Discovery Framework Work?

• AI Design Phase: NYCU's Drug Design and Systems Biology Laboratory uses screening models to generate antibody sequences targeting specific disease proteins, prioritizing candidates with higher potential based on computational analysis.
• Experimental Validation: HanchorBio's biologics engineering team tests the AI-designed candidates in the lab, assessing their binding affinity, functional activity, and development potential through rigorous experimental protocols.
• Data Feedback Loop: Results from experimental testing are fed back into the AI models, allowing them to learn from real-world performance and continuously improve predictions for future targets and programs.

This iterative approach differs fundamentally from one-directional AI tools that generate designs without learning from experimental outcomes. The closed-loop model means each round of testing makes the AI smarter, potentially accelerating discovery for subsequent programs.

The initial focus targets antibody discovery, but the collaborators envision expanding the framework across multiple disease targets and biologics programs over time. HanchorBio specializes in immunotherapies for oncology and autoimmune diseases, so the partnership aims to support development of next-generation molecules that can activate both innate and adaptive immune pathways.

"The AI technology used in this collaboration is based on screening models developed by our Drug Design and Systems Biology Laboratory. Our goal is to use data-driven modeling to prioritize higher-potential candidates earlier in the discovery process, while maintaining rigorous experimental validation of their properties and functional activity," explained Jinn-Moon Yang, Institute of Bioinformatics and Systems Biology at National Yang Ming Chiao Tung University.

Jinn-Moon Yang, Institute of Bioinformatics and Systems Biology, National Yang Ming Chiao Tung University

What Makes This Partnership Different From Existing AI Drug Discovery Approaches?

Many AI drug discovery platforms focus on computational prediction alone, generating thousands of candidate molecules without experimental validation. Others rely on traditional screening methods with minimal AI integration. The HanchorBio-NYCU model sits in the middle, using AI to intelligently narrow the search space before committing resources to experimental testing.

This hybrid approach addresses a real bottleneck in modern drug discovery. Computational models can predict molecular properties with reasonable accuracy, but they often miss nuances that only emerge in real laboratory conditions. By validating AI predictions experimentally and feeding results back into the models, the collaboration creates a system that becomes more accurate and reliable over time.

The partnership also reflects a broader industry trend toward integrating AI into existing biologics platforms rather than replacing traditional discovery entirely. HanchorBio brings its proprietary FBDB (Fc-Based Designer Biologics) platform and translational expertise, while NYCU contributes advanced AI modeling capabilities developed by Professor Jinn-Moon Yang's laboratory, which specializes in applying artificial intelligence and systems biology to understand relationships among drugs, proteins, and disease mechanisms.

What Are the Practical Implications for Drug Development Timelines?

If successful, this collaboration could meaningfully compress early-stage discovery timelines. By reducing the number of candidates requiring experimental screening and improving the quality of those that do advance, researchers may move from target identification to lead candidate selection faster than conventional workflows allow.

The impact extends beyond speed. Better candidate selection earlier means fewer resources wasted on molecules unlikely to succeed in development. This efficiency gain could lower discovery costs and allow biotech companies to pursue more ambitious targets or expand their pipelines without proportional increases in spending.

For patients, faster and more efficient discovery translates to new treatment options reaching clinical testing sooner. Immunotherapies for oncology and autoimmune diseases remain areas of significant unmet medical need, so accelerating the path from lab to clinic could have meaningful health implications.