Drug Discovery Has No Magic Wands
The article argues that AI’s success in molecular design won’t cure most diseases without a deeper understanding of human biology and disease mechanisms. It calls for large-scale, causal biological measurements and better links between cellular data and clinical outcomes before AI can deliver truly transformative medicines.
- AI drug discovery has focused on molecule design, but the real bottleneck is upstream: most diseases lack a validated biological target to hit.
- 90% of clinical trials fail because they target the wrong biological mechanism, not because molecule design is too slow.
- Genuinely new drug targets have dropped from ~100 in 2015 to ~30 in 2024, while 38 targets each now have 50+ drug programs piling onto the same well-known locks.
- Fixing this requires massive causal/perturbational biology data (billions more measured cell states), not just better AI models, since current cell atlases barely scratch the surface.
Daphne Koller argues that AI in drug discovery won’t pay off until we vastly improve our understanding of human biology. She breaks drug development into three stages: finding disease mechanisms, designing molecules to hit those targets, and running clinical trials. AI work so far has zeroed in on molecule design, inspired by successes like AlphaFold. But no AI tool has yet “drugged the undruggable.” The real bottleneck lies upstream: most diseases lack a validated mechanism to target, so faster molecular design simply produces more failures.
Koller points out that 90% of clinical trials fail because they aim at the wrong biological locks. We’ve thrown resources at a handful of well-studied targets—38 of them now have more than 50 drug programs each—while the number of genuinely new targets has plunged from about 100 in 2015 to roughly 30 in 2024. Patients suffer because we keep refining keys for familiar locks instead of mapping the unknown parts of the system. Expanding modality toolkits—antibodies, siRNA, gene editing—helped before, but without fresh mechanistic insight, we’re back to square one.
She calls for a massive push to measure biology at scale, especially causal and perturbational data. Cell atlases cover only a tiny slice of the cellular universe, and they lack most of the data on how cells respond to interventions. To turn AI into a real discovery engine, we need experiments that sample billions more cell states under varied conditions. Only with that data foundation can AI reasoning shift from generating failures faster to finding the right therapeutic paths.