Finding the molecular switches behind new infectious diseases

Clare Bryant studies why viruses that jump from animals to humans, like flu or Ebola, sometimes trigger deadly overreactions in our bodies such as sepsis. It's slow, grinding work: normally you spend two or three years testing proteins one by one before narrowing in on the actual amino acids responsible. So when Google DeepMind offered her a shot at their research tool, Co-Scientist, she treated it like any skeptical scientist would — as an experiment to be tested, not trusted.

Her first try was modest. She fed the AI a summary of a grant proposal about flu in birds and humans, and it spat back a ranked list of hypotheses. Some she'd already thought of. A few she hadn't, and those were the ones that stuck with her.

Things got more interesting once the grant actually landed and she gave Co-Scientist the full proposal instead of a summary. Reading the output on a train to Brussels, she spotted a protein the tool had flagged that she hadn't been tracking, one that tied neatly into signaling pathways she already cared about. She described spending the rest of that week wanting to feed it everything she had.

Back in the lab, she did exactly that, feeding in unpublished data kept private within the tool. With each round of back-and-forth, the hypotheses got sharper, moving from vague protein candidates down to specific amino acids worth testing. Her team is now building cell lines with those exact mutations to check whether the AI actually pointed them somewhere useful.

If it holds up, Bryant expects to hit in six months what would normally take two to three years. That's not a small claim in a field where grant cycles and PhD timelines are built around exactly that kind of delay.