The AI system Astra produced five draft applications for multiyear US National Institutes of Health grants in one hour
The AI system Astra produced five draft applications for multiyear US National Institutes of Health grants in one hour
On September 6, Locasale wrote on X that he had asked Astra to read his published work, identify new research directions, and prepare five R01 applications for the US National Institutes of Health. An R01 is a multiyear grant for biomedical research. According to Locasale, the work took about an hour and approximately 20 dollars in computing costs.
The attached page from one draft addresses a question about cancer cell metabolism. The response of these cells to enzyme inhibition may change when the available nutrients change. The draft aims to determine which enzyme controls the rate of the process under these conditions.
The draft uses the laboratory’s previous findings as the basis for a new question. Its hypothesis is that a cell’s nutrient supply, chemical state, and energy requirements determine which step in glucose metabolism controls the rate of the process. The first aim proposes varying these conditions and the activity of three enzymes, then testing which step becomes rate-limiting.
The second aim proposes comparing several explanations with experimental measurements and constructing a simplified model that accounts for the observed changes. The third aim requires specifying predictions in advance for new combinations of nutrients and new cell cultures used in the experiments, then comparing those predictions with the experimental results. This single page therefore shows how previous observations can be developed into a testable experimental plan.
Locasale wrote that Astra “reproduced this logic well” and turned existing research into questions, hypotheses, and experiments that a review panel would find familiar and well supported. He described all five drafts as “quite reasonable,” grounded in his work, and similar to applications he might have written himself.