The panel unanimously agrees that AI has delivered reliable, significant value in early-stage drug discovery, such as protein design and molecule generation. However, its ability to predict success in Phase II and III clinical trials remains poor due to biological complexity and data limitations, creating an expensive 'last mile' problem.
A recurring point is that the predictive power of AI is fundamentally limited by the data it's trained on. While pre-clinical data is more abundant, clinical trial data is often siloed, sparse, and not longitudinal, preventing AI from effectively modeling complex human disease progression and therapeutic response.
AI and machine learning are enabling a paradigm shift, moving biology from a qualitative, descriptive science to a quantitative, engineerable discipline with mathematical underpinnings. This transition from 'pictures to formulas' is compared to industrial revolutions in other fields like physics and electricity.
AI is poised to fundamentally change the nature of scientific work by automating hypothesis generation and testing at an unprecedented scale. Instead of a scientist testing one hypothesis over six months, AI agents can now test 10,000 hypotheses in ten days, shifting the human role from execution to synthesis and strategy.
The panelists touch upon the need for regulatory bodies like the FDA and the clinical trial system itself to evolve. Current trial structures (Phase I-III) and statistical models may be outdated, and regulators may lack the AI expertise to evaluate novel, AI-driven submissions and adaptive trial designs.
Keep pulling the thread on Athos Therapeutics.