Generative AI and large language models alone cannot solve the hardest problems in drug discovery, Hajo Schiewe, VP Strategic Business Development & Scientific Partnerships at SandboxAQ, argues in his DLD Health talk. These systems “don’t work very well on tasks where the data isn’t present in the data set”, he says. “So if you have a new molecule for a disease which hasn’t been solved before, it struggles with assessing that molecule.”
As an alternative, he introduces large quantitative models (LQMs), “models which reason from physics and not from past patterns”. These AI models are trained on simulated data derived from first principles, he explains, and can be applied to a wide range of tasks. “The territory of LQMs ranges from protein folding, drug target binding, mRNA design [to] antibody optimization.”
The goal is not to replace large language models (LLMs), Schiewe stresses, but to combine them with LQMs. LLMs enable intuitive interaction, while LQMs solve “complex quantitative tasks”, he says. This integration also expands access: users can trigger advanced simulations through familiar interfaces without specialized expertise.
Watch the video to explore the talk in detail.
