A computational method combining generative AI with atomistic simulations can identify promising platinum alloy catalyst structures for hydrogen fuel cells, report researchers from Science Tokyo. Their approach addresses a longstanding challenge in catalyst design and consistently produces high-performing candidates from several material combinations.

CAMBRIDGE, Mass. - April 29, 2026. Researchers at the Massachusetts Institute of Technology have introduced Bioinspired123D, a generative AI system that translates plain-English design descriptions into executable 3D geometries inspired by biological materials, from crab exoskeletons to horse hoof walls. The work, published in AI for Science, shows that controllable, high-fidelity 3D generation for scientific design does not require massive 3D foundation models.

Developed by graduate student Rachel K. Luu and Professor Markus J. Buehler in MIT's Laboratory for Atomistic and Molecular Mechanics (LAMM), the system departs from mainstream text-to-3D methods that rely on meshes, voxels, or point clouds. Instead, Bioinspired123D generates compact Blender Python scripts — programs that, when executed, produce parametric, smooth, and fabricable structures. This "code-as-geometry" representation makes the resulting designs interpretable, editable, and ready for downstream simulation or 3D printing.

"Biological materials encode an extraordinary amount of design intelligence in their geometry - helicoidal plies in stomatopod clubs, gradient porosity in horse hooves, cellular cores in bird beaks," said Buehler, the McAfee Professor of Engineering at MIT and senior author of the paper. "We wanted to build a system that can take that intuition, expressed in natural language, and turn it directly into structures you can fabricate, without needing a supercomputer to do it."

Implanting a device into the deep temporal cortex of a mouse without damaging the brain has long been a major challenge in neuroscience research. A team at Meijo University and Dokkyo Medical University has now overcome this barrier with a flexible sheet thinner than a human hair that slides into place without penetrating the brain.

Biomimetic Strategy Simultaneously Enhances Output, Mechanical Strength, and Transparency; Extended to Wearable Motion Recognition Sensor Combined with Machine Learning

For more than a decade, a fundamental mystery has surrounded graphene—the one-atom-thick “wonder material” known for its exceptional strength, conductivity, and transparency. Despite its seemingly simple structure, one basic question has remained unresolved: does graphene attract water, or repel it?

The answer has proven surprisingly elusive. In some experiments, water droplets bead up on graphene, suggesting a hydrophobic (water-repellent) surface. In others, water spreads out, implying hydrophilic (water-attracting) behavior. This contradiction has fueled a long-running scientific debate and created uncertainty for applications such as desalination membranes, hydrogen fuel cells, and nanoelectronic devices, where precise control of water at interfaces is essential.

A research team led by Director CHO Minhaeng and Professor Stefan RINGE at the Center for Molecular Spectroscopy and Dynamics within the Institute for Basic Science, in collaboration with Korea University, has now resolved this puzzle. Using machine-learning–enhanced molecular simulations, the researchers demonstrate that pristine graphene is intrinsically hydrophobic and microscopically not wetting transparent.