Agentic AI for Drug Discovery with
Physical and Biological Priors
Artificial intelligence is reshaping how therapeutic targets are identified and how chemical and biological design space is explored. Drawing on work at AIGEN Sciences, this talk describes an AI-driven discovery framework in which LLM-based agents coordinate domain-specific foundation models across target prioritization, hypothesis generation, molecular design, and iterative optimization.
A recurring theme is that structure prediction is becoming commoditized infrastructure, while experimental labels linking a designed molecule to its function remain scarce for the modalities of greatest current interest. Progress therefore depends on encoding physical and biological priors directly into the generative process rather than applying them as post hoc filters.
We illustrate this across three modalities — molecular glue degraders, therapeutic peptides, and antibodies — each of which requires different structural and biological constraints to be respected during generation. We close with the problems that remain hardest: the scarcity of measured binding and degradation labels, the gap between computational energy proxies and
experimental affinity, and the accuracy limits of complex structure prediction.
Jaewoo Kang, Ph.D.
Korea University
Jaewoo Kang is a Professor in the Department of Computer Science and Engineering at Korea University and the Founder and CEO of AIGEN Sciences, an AI-driven drug discovery company. His research sits at the intersection of artificial intelligence and biomedicine, spanning foundation models, generative molecular design, and multi-modal learning for drug discovery and precision medicine.
Professor Kang is best known for BioBERT, one of the first domain-specific language models for biomedical text and now among the most widely adopted foundation models in the field. His current work centers on agentic AI systems that couple foundation models with computational and experimental workflows, closing the loop from target identification through lead discovery and optimization. At AIGEN Sciences, the same approach is being applied to emerging therapeutic modalities such as molecular glue degraders, where generative design is paired with ternary complex prediction and degradation activity modeling.
His work has been cited over 28,000 times, and his teams have won ten international biomedical AI competitions, including the DREAM Challenges and the BioASQ challenge. Across academia and industry, he is committed to building AI systems that act as genuine scientific collaborators, shortening the path to therapies for diseases with unmet medical need.