Uncovering repurposed medicines to fight liver fibrosis

Liver fibrosis kills more than 1.4 million people a year through the cirrhosis it eventually causes, and there's still no drug specifically approved to stop it. So when Stanford geneticist Gary Peltz decided to go looking for repurposed medicines that might help, he brought in some unusual help: Google DeepMind's Co-Scientist AI tool.

Peltz ran a simple head-to-head test. He picked two candidate drugs himself, based on how often they showed up in the liver fibrosis literature. Co-Scientist, meanwhile, proposed three of its own picks, along with its reasoning for each. All five went into Peltz's fibrosis testbed, a system built from live human liver cells that can show whether a compound actually slows scarring or helps cells regenerate.

The human picks flopped. Neither of Peltz's own choices showed any benefit against fibrosis in the lab. Co-Scientist's suggestions did much better: two of its three candidates blocked fibrosis and encouraged liver cell regeneration. One of those drugs had barely registered in the fibrosis literature at all, just a handful of papers, the kind of obscure connection that's easy for a person to miss and exactly the sort of thing an AI system sifting through mountains of text is built to catch.

The standout was vorinostat, a drug currently used to treat certain cancers. In Peltz's experiments it blocked 91% of the damage response thought to drive liver scarring, a striking number for a compound nobody had been studying for this purpose. What ties Co-Scientist's picks together, according to Peltz, is that they don't hit a single fibrosis pathway the way most drug candidates do. Instead they reshape broader patterns of gene activity, a strategy that's harder for a human researcher to reason toward but apparently well within reach for a machine reading across the entire scientific record.

Peltz now argues that this class of gene-activity-altering drugs deserves serious attention as a starting point for a new generation of anti-fibrotic medicines, a shift in strategy nudged along not by a hunch, but by an AI that read more papers than any one scientist ever could.