These three things are the same story. The grant US10647980B2 (issued May 12, 2020) describes a platform that designs, builds, and tests microbial strains at high throughput. The method is data-driven optimization over an experimental loop. And the framing, software steering wet-lab experiments, is exactly what the market would later brand 'AI for science.'
Connect the dots and the pattern is older than the hype. Long before foundation models could propose protein structures, companies were filing IP on the closed loop of propose-experiment-measure-update. The intelligence in 2020's version lived more in the search and statistics than in a giant neural network, but the architecture of the idea is identical.
Follow both the money and the IP and you see why this matters for the sector narrative. The 'AI discovers a drug' headline of 2024 rests on infrastructure patented years earlier: automated labs, high-throughput assays, and the optimization layer that turns measurements into the next experiment. The grant is the receipt for that groundwork.
The honest caveat: this is not a transformer, and calling it 'AI' is a stretch by 2026 standards. But the lineage is real. The reason AI-for-science scaled so fast is that the laboratory automation and the data pipelines were already built and protected. The model was the last piece to arrive, not the first.
For a reader tracking where AI actually creates value, the lesson is to look past the model to the loop it sits inside. The grant tells you the loop existed in 2020. The models just made the proposing step a lot smarter.