Understanding cancer at a genetic level with AI

The detail worth sitting with is the hardware constraint: Makerere's team ran this screening workload on commodity machines, which means the bottleneck DeepMind is quietly dissolving isn't just compute access but the assumption that frontier biomedical AI requires frontier infrastructure budgets.

This fits directly alongside the 'outsmart drug-resistant bacteria' story from the same day, where DeepMind researchers at Cambridge used AlphaFold and Gemini to compress drug discovery timelines. Taken together, those two pieces sketch a consistent pattern: DeepMind is positioning AlphaFold not as a research artifact but as a reusable substrate for applied clinical problems across very different institutional contexts. The Co-Scientist coverage from the same date adds a third data point, showing DeepMind pushing AI into the hypothesis-generation layer of science, not just the validation layer. What's emerging is a layered biology stack, with AlphaFold handling structure, AlphaGenome handling regulatory genomics, and Co-Scientist potentially sitting above both as a reasoning interface.

Watch whether Makerere or a comparable Global South institution publishes wet-lab validation results on any of those 15 candidates within 18 months. If even one advances to preclinical testing, it becomes a concrete proof point that this workflow produces actionable biology rather than a well-funded case study.