How to See a Molecule's Shape Under a Microscope express
A scanning tunneling microscope probes a molecule with an ultra-fine tip and maps its relief. But the image is flat, while the molecule is three-dimensional. How do you guess a person's pose from their shadow on a wall? Scientists built a "translator from shadows to volume"—a neural network that reconstructs the 3D shape of a molecule from its imprint.
First, the network was trained on millions of synthetic images. The computer calculated how a molecule would look under the microscope if it lay this way or that. Then they showed it a real image, and the network proposed the most likely pose. For the peptide bradykinin, the error was less than two angstroms — ten times thinner than a human hair.
The method was also tested on sugars — glycans. They are more flexible than proteins, so the error is slightly larger — up to four angstroms. But even so, the computer correctly found the positions of rings and bonds. This is no longer just a pretty picture, but a working tool for chemists and biologists.
Now determining a molecule's structure takes minutes, not weeks. You can quickly check how a drug fits its target or how a mutation changes a protein's shape. The shadow on the wall has spoken.
Original abstract
as written by the authors arXiv:2603.07706
Improving the detailed understanding of the underlying properties and functions of biomolecules has recently attracted growing interest, enabled by the possibility of real-space imaging of single, intact macromolecules using Scanning Tunneling Microscopy (STM) in combination with electrospray ion beam deposition and soft landing. This combination provides key insights into biomolecular behavior, but it also imposes stringent requirements on rapid and reliable data analysis. A major limiting factor for applying machine learning to STM images is often the scarcity of training data, caused by the long acquisition times required for both experimental imaging and high-accuracy simulations. Here, we propose a framework for the rapid generation of three-dimensional structures of glycans, peptides, and glycopeptides and their corresponding STM-like image simulations, based on state-of-the-art, machine-learning-accelerated Density Functional Theory (DFT). We generate datasets for the polypeptide bradykinin and for a representative glycan molecule, and we train a conformer estimation model to predict a molecule's three-dimensional structure from an STM image. On synthetic data, our approach achieves high accuracy, with median atomic deviations below $$2\,\AA{}$$ for peptides and below $$4\,\AA{}$$ for glycans. Application to experimental data predominantly yields a precise, reliable, and visually convincing determination of the local positions of molecular subunits. The application to experimental data represents an important milestone towards a fully automated structural search pipeline for complex, biologically relevant systems imaged with STM.
An angstrom is one ten-billionth of a meter. The diameter of a hydrogen atom is about one angstrom.
In the series "Star Trek," the medical tricorder identified the composition of matter in seconds. Our method is a small step toward such a scanner.