The square peg that reshapes the hole: a week of science bending the impossible

A square peg has met its round hole—and the hole is the one that changed shape. Researchers at the University at Buffalo built a drug molecule that doesn't fit its target protein, p38 delta, implicated in cancer. Instead of failing, the molecule forced the protein's flexible loop to bend and wrap around it, creating a snug fit so exact it represented a 12,000-fold improvement in selectivity and was 110-fold more potent than anything previous.

"It's typically you design a molecule to fit the target," says corresponding author David Heppner. "Here, we designed a molecule that makes the target adapt to it."

That stubborn refusal to follow the old rules is the through-line running through eight studies published this past week—each one quietly rewriting how science builds, measures, and understands the world.

The pangenome that could save the ash

In the UK, a silent catastrophe is unfolding in the woodland. Ash dieback has been devastating European ash trees, but a new weapon has arrived. Researchers from the Royal Botanic Gardens, Kew and Forest Research published in Nature Communications the first-ever European ash (Fraxinus excelsior) pangenome, constructed from 50 trees of diverse origins.

By analyzing a genomic dataset from more than 1,000 trees, they identified 211 genes possibly linked to ash dieback resistance—16 of them "dispensable," present in some trees but not all. A traditional genome assembly captures only a single individual, which is why dispensable genes went unseen. The team found more than 3,400 such genes, just 9% of the total, that may help populations adapt quickly to disease or drought.

A game of 20 questions for cheaper AI

At the University of Bristol, professor Sidharth Jaggi and colleagues are slashing the cost of training AI with a trick from childhood. Instead of running tens of thousands of graphics processing units that cost millions of dollars, they combine simple binary classifiers—each trained in minutes on a standard laptop—using a string of yes-or-no questions, like 20 Questions finding an unusual species mid-walk.

"We are able to mathematically prove—and back up by empirical validation—that even if we ask 'simple random questions,' the answers can be combined to perform complex tasks," Jaggi says. The work was presented at the Allerton Conference in Illinois.

Catalysts, DNA skyscrapers, and a resilient receptor

At University of Michigan Engineering, a stability rulebook for dilute alloy catalysts was published in the Journal of the American Chemical Society—materials made of 1% or less of an active dopant metal that could slash precious-metal use in fuel, plastic, and pharmaceutical production. And at Sungkyunkwan University in Korea, Do Hyun Ryu's team developed two new asymmetric catalytic methods published in Angewandte Chemie International Edition, controlling multiple reaction sites and stereocenters at once.

Meanwhile, 44 years after Nadrian Seeman first imagined DNA as a building material, William Shih's group at the Wyss Institute and Dana-Farber Cancer Institute has given DNA origami a software upgrade—self-assembly methods that build stable, rigid 3D multilayered structures for nanorobots, targeted drug delivery, and even the DoriVac project, which organizes vaccine components at the nanoscale to fight cancer.

And in Tanzania, a 240-million-year-old fossil named Dinodontosaurus isiyavamanda, described by a University of Bristol-led team in the Journal of Vertebrate Paleontology, is refining the age of one of the world's most significant fossil sites. Finally, the University of Osaka's team mapped the 3D structure of a pufferfish umami receptor in PNAS—finding it detects a far wider range of amino acids than its human counterpart.

The thread that binds them

What connects a protein that bends, a pangenome, a cheap AI trick, and a fish's tongue? None of them accept the structure as fixed. Each team looked at what was previously "impossible"—a loose loop, a rigid genome, a million-dollar compute bill—and found a more flexible path forward. In the search for ash trees that survive, drugs that outsmart cancer, and AI that fits in a laptop, the scientists of this week prove the same lesson the pufferfish teaches its human observers: sometimes the strongest adaptation is versatility itself.