Monday, June 15, 2026

On the trail of the missing hydrogen atoms

Paul Scherrer Institute

Timo Reents and Giovanni Pizzi — image:
Timo Reents (left) and Giovanni Pizzi have taught an artificial intelligence system to find missing positions of hydrogen atoms in crystal structures.
view more
Credit: © Paul Scherrer Institute PSI/Mahir Dzambegovic

Artificial intelligence is often used to generate images. In research, specialised AI models are used for scientific applications – for example, to predict the positions of atoms in materials. The MatterGen model developed by Microsoft can generate complex crystal structures from just a few pieces of information – which atoms should be present and in what proportions – and researchers can then use these structures for computer simulations of new materials.

Now a scientific team led by Giovanni Pizzi from the PSI Center for Scientific Computing, Theory and Data, together with researchers from the universities of Parma and Modena in Italy, has found a way to use AI to solve a practical problem in materials science: locating missing atomic positions in otherwise known structures. As they report in the journal npj Computational Materials, the materials scientists used an approach normally employed in image processing or computer vision, that is, recognition and interpretation of visual information by means of AI.

This allows materials that are experimentally known but have been theoretically inaccessible to be simulated for the first time or significantly better than before. Thus the researchers are contributing to the exploration of new materials with special properties, for hydrogen storage for example, or potentially for the development of new superconductors.

“Invisible” hydrogen atoms

“For our simulations of material properties, we rely on information in databases telling us where each atom is located in a crystal structure,” says Timo Reents, a doctoral candidate in Giovanni Pizzi’s group. However, the element hydrogen presents a challenge. It is often part of the crystal lattice, but it is difficult to detect experimentally using traditional methods that measure the arrangement of atoms through X-ray diffraction. Consequently, the positions of hydrogen atoms in crystal representations are often inaccurate, or they are missing altogether from the visualisations.

Precise knowledge of the atomic positions is essential for computer simulations that researchers use to predict specific material properties, such as electrical or thermal conductivity. “If the information about the hydrogen atoms is missing, that’s a problem,” says Giovanni Pizzi. “Often, we can’t use several thousand potentially interesting materials for our simulations precisely for this reason.” This is where AI should be able to help.

When a paw is missing from a photo of a dog

In computer vision, so-called diffusion models are used. When these are used to fill in missing image information, it is called inpainting. For example, a paw that was hidden in a photo of a dog can be added.

Earlier approaches to computer vision would first add “noise” to the entire image of the dog, intentionally overlaying it with random image information, in order to then reconstruct the photo with all four paws in a second step. Now, however, it is standard practice to vary the strength of the noise depending on the image area: Noise would be added heavily only to the unknown regions where the paw should be.

While this is already established in the field of computer vision, it was previously unavailable for the reconstruction of atomic positions. Now Giovanni Pizzi’s team has developed an adapted open-source model called XtalPaint, based on Microsoft’s MatterGen. “This combines the advantages of modern computer vision and crystal reconstruction: Noise is added only to the unknown positions within the crystal – the known positions remain largely unchanged during the process,” Timo Reents explains.

This offers greater efficiency, just as it does in modern inpainting approaches in computer vision: “With step-by-step reconstruction, XtalPaint can orient itself to the existing crystal from the very beginning,” Reents says. “This increases the success rate and also saves computing power.”

Also applicable to lithium and sodium

To test their method, the researchers removed the hydrogen atom positions from known crystal structures and then used XtalPaint to reconstruct them. In 87 percent of cases, they found the known positions – and in another ten percent, configurations that were even more energetically stable. “Overall, this means a success rate of 97 percent for XtalPaint,” Reents says.

“We can now use our method, for example, to complete structures in databases with the missing hydrogen positions,” says Pizzi. Also, he and his colleagues have already detected errors in databases that can arise through data transfer from original scientific publications. Furthermore, they can apply the method not only to hydrogen atoms, but also for example to lithium and sodium – two elements that are important for the development of new batteries.

Text: Oliver Morsch

In the inpainting method, the artificial intelligence system is trained to preserve a known crystal structure (blue, black, and red spheres) and only insert the missing hydrogen atoms (blurred on the left, white spheres on the right).

Credit

About PSI

The Paul Scherrer Institute PSI develops, builds and operates large, complex research facilities and makes them available to the national and international research community. The institute's own key research priorities are in the fields of future technologies, energy and climate, health innovation and fundamentals of nature. PSI is committed to the training of future generations. Therefore about one quarter of our staff are post-docs, post-graduates or apprentices. Altogether PSI employs 2300 people, thus being the largest research institute in Switzerland. The annual budget amounts to approximately CHF 450 million. PSI is part of the ETH Domain, with the other members being the two Swiss Federal Institutes of Technology, ETH Zurich and EPFL Lausanne, as well as Eawag (Swiss Federal Institute of Aquatic Science and Technology), Empa (Swiss Federal Laboratories for Materials Science and Technology) and WSL (Swiss Federal Institute for Forest, Snow and Landscape Research).

Journal

npj Computational Materials

DOI

10.1038/s41524-026-02090-1

Method of Research

Computational simulation/modeling

Subject of Research

Not applicable

Article Title

Score-based diffusion models for accurate crystal-structure inpainting and reconstruction of hydrogen positions

Article Publication Date

11-Jun-2026

An electron microscope breakthrough delivers sharper images of our body’s tiniest proteins

Physicists introduce phase contrast to electron microscopy, potentially revolutionizing cryo-EM

University of California - Berkeley

Cryo-EM with laser phase plate — image:
A laser (purple) is powerfully amplified by highly polished mirrors and focused on the electron beam (blue) to shift its phase and increase the cryo-EM microscope’s contrast, allowing biologists to image smaller proteins and the crowded structures inside cells.
view more
Credit: Sayo Studio

Nearly 100 years ago, a seemingly simple discovery revolutionized the microscope. The introduction of phase-contrast, which garnered a Nobel Prize in 1953, brought into clear view structures inside cells that had previously been too faint or washed out for biologists to study.

UC Berkeley physicists have now adapted the phase contrast technique to the electron microscope, which has about 10,000 times the magnification of microscopes using optical light.

The addition of a so-called laser phase plate has the potential to greatly improve cryoelectron microscopy (cryo-EM), a technique for determining the structure of molecules that itself revolutionized the understanding of proteins and accelerated new drug discovery starting a decade ago. Despite its impact, however, cryo-EM still struggles to produce clear images of small molecules — including most human proteins. A laser phase plate promises clear images of most proteins in the cell down to one-third the size of those that are a challenge for today’s machines.

The addition of a laser phase plate seems certain to revolutionize a newer technique referred to as cryoelectron tomography (cryo-ET), which assembles a number of different angular views of a molecule or protein into a 3D image. This makes it possible to analyze proteins in their natural environment — inside cells — instead of in isolation in a solution.

“Cryo-EM has become the new, fastest-growing method for resolving the structure of biological macromolecules, and cryo-ET is expected to show how these molecules work together in their natural, cellular context,” said Holger Müller, a UC Berkeley professor of physics and faculty scientist at Lawrence Berkeley National Laboratory who led the development effort. “But because of signal-to-noise limitations, the majority of human and animal proteins are too small to be analyzed by these methods. The increase in signal-to-noise ratio provided by this laser phase plate is expected to overcome these important limitations.”

Crucial to the development is the world’s most intense, focused continuous-wave laser, which interacts with the electron beam to change its phase. This phase change boosts contrast for small molecules, such as hemoglobin, and for molecules and structures inside cells, such as the nucleus and mitochondria.

“With cryo-ET, we're looking at small, very complicated cellular material that’s incredibly crowded inside the cell,” said Bridget Carragher, founding technical director of imaging at Biohub in Redwood City, California “It’s like a forest of trees, and you’re trying to find one leaf on one tree in there. Cryo-ET needs a dramatic step forward in contrast, so we can start to see what's going on inside the cell. That's what the laser phase plate promises to give us.”

Biohub provided funding to Müller to purchase a state-of-the-art cryo-EM machine that he then outfitted with a laser phase plate, creating a microscope he calls Theia, named after the ancient Greek Titaness of light and radiance. Carragher is overseeing the development of a similar instrument at Biohub’s imaging lab in Redwood City — this one featuring a dual-laser system, based on theoretical work by Müller and his colleagues. In this system, the two perpendicular laser beams operate at about half power, making the components less likely to burn out and reducing aberrations.

Both groups are collaborating with the firm Thermo Fisher Scientific, the primary manufacturer of cryo-EM machines.

“Theia is the Formula 1 microscope,” Müller said. “It has extra electron optics that give it better resolution than the standard cryo-EM, even without the laser. With the addition of the laser phase plate, we hope that it really becomes the world's best instrument overall.”

Müller and his Berkeley team will publish their newest images and details of the cryo-EM’s laser phase plate in the June 11 issue of the journal Science. Biohub’s two-laser system is described in a recently posted preprint.

Biological imaging

Animal and plant cells are mostly water and thus transparent in a light microscope, which should make it easy to see structures such as the nucleus and mitochondria inside. But these structures are small and scatter only a small amount of light, which makes them only slightly darker than the rest of the cell’s insides. This low contrast has typically been improved by staining the cell, though staining also kills the cell.

In 1930, Dutch scientist Frits Zernike realized that the brightness or amplitude of the light was not the only feature affected when passing through a cell. The scattered light is also slowed down in a biological sample, which shifts its phase — the timing of the peak of the waveform — by a small amount. While this phase shift is invisible to the human eye, it can be turned into visible contrast by also phase shifting the non-scattered light by 90 degrees. When the scattered and non-scattered light are ultimately focused on the retina, features in the sample are enhanced relative to the background, boosting the contrast. Zernike received the 1953 Nobel Prize in Physics for this discovery.

By the early 1940s, the phase-contrast microscope had proved its value and scientists speculated about adapting this technique to increase contrast in the electron microscope, which uses a beam of electrons to image much smaller structures, such as proteins. But attempts to make a phase plate that shifts the phase of an electron beam reduced the beam intensity too much, made the images unstable, or resulted in lower resolution.

In 2010, Müller and Robert Glaeser, now a Berkeley professor emeritus of molecular and cell biology, wrote a paper proposing a way to create the phase shift by using an intense laser, which would not dim the electron beam.

Glaeser is a pioneer of cryo-EM, a major improvement in electron microscopy and theoretically a simpler method for determining molecular structures than X-ray crystallography, which requires that a molecule actually forms a crystal and that the researcher has access to a bright source of X-rays. But a major problem with electron microscopy is that the electron beam heats up and eventually damages its target, limiting image detail. Coating the specimen with metal to prevent this and enhance the contrast only makes fuzzier images.

In the 1960s, scientists proposed freezing samples to slow down sample destruction. Glaeser demonstrated that freezing samples reduced radiation damage and proposed reducing damage even further by lowering the power of the electron beam while irradiating thousands of frozen molecules simultaneously. Though each molecule in the sample would be in a random orientation, computers could combine all the images to create a highly detailed structure.

The originators of cryo-EM were awarded a Nobel Prize in Chemistry in 2017, and in their acceptance remarks credited Glaeser’s work. According to the Nobel Committee, cryo-EM “both simplifies and improves the imaging of biomolecules. This method has moved biochemistry into a new era.”

After the publication of the 2010 paper, Müller spent 15 years realizing the goal of a laser phase plate for cryo-EM, funded in part by a grant from the National Institutes of Health. First, he and his team had to develop a way to focus a continuous laser onto a small spot to create light intense enough to shift the phase of an electron beam by 90 degrees. After 10 years, they achieved this by trapping the laser beam in a spherical, mirrored cavity that both focuses the beam and intensifies it as the light bounces back and forth more than 10,000 times.

“It’s 75 kilowatts focused to a few microns,” Müller said. “That’s more powerful than what you use for welding. It's more power than a military laser. It builds up the brightest continuous laser focus ever.”

They proved that the concept worked by installing a laser phase plate in one of Glaeser’s old microscopes, but Biohub funding later allowed them to purchase a customized, state-of-the-art Thermo Fisher Krios cryo-EM microscope and refit it. In the new paper, they demonstrate that the powerful focused laser beam produces higher resolution images for six different samples of different sizes and different sample preparation.

“For the most challenging cases — small particles, bad specimens — the laser produces a very considerable advantage,” Müller said.

In their paper, they show reconstructed images of a protein from muscle called aldolase, which is relatively easy to image with today’s cryo-EM machines, and for hemoglobin — a protein that carries oxygen in blood — which is at the lower limit for current machines. The laser phase plate improved the resolution of the protein structure in both cases, but more so for the smaller molecule, hemoglobin.

“The bottom line is, if you have a large protein and a really good sample — a fresh one or one frozen without bubbles, for example — you may not need the phase plate to get a single, high-quality image. But for a small protein and a bad sample, laser-on is best,” Müller said. “This could fill an enormous gap in our knowledge of protein structures that can’t be crystallized or are too small for today’s cryo-EM. And it will be revolutionary for cryo-ET.”

Protein size is measured in daltons — named after English chemist John Dalton and equivalent to 1/12 the mass of a carbon-12 atom — and cryo-EM today can barely image proteins smaller than 70 kilodaltons, which make up about 90% of the human proteome. With the laser phase plate, it’s now possible — though difficult — to image down to 50 kilodaltons (even smaller than hemoglobin).

Soon, Müller hopes, this will be improved to 17 kilodaltons (the size of the protein myoglobin). He is optimistic that that can be achieved with a focused electron beam, as opposed to a defocused beam, which without the laser phase plate is now required to get any contrast at all. This advantage would be another benefit of the laser phase plate and would deliver another factor-of-two boost in contrast and signal-to-noise ratio, on top of the one already achieved. A laser phase plate should be able to extract contrast from phase changes in the focused electron beam alone, he said.

“This technology is a step function change for biology,” said Stephani Otte, Biohub’s Vice President of Imaging Science. “We are going to be able to see how molecular machines operate inside the living cell, in context, for the first time. What was once invisible will become visible — and that changes everything about how we understand disease.”

Müller’s co-authors are Glaeser, UC Berkeley postdoctoral fellows and co-first authors Petar Petrov and Jessie Zhang; staff scientist Jonathan Remis, postdoctoral fellow Hang Cheng; and current and former physics graduate students Jeremy Axelrod, Eric Cooper, Ian Hicklin, Shahar Sandhaus and Cooper Schnurr.

Carragher and David Agard, founding scientific director of imaging at Biohub, are co-leads of Biohub’s Dynamic Structural Cell Biology group and co-corresponding authors of the Biohub preprint, along with Biohub engineer Pavel Olshin.

Cryo-EM protein images with and without laser phase plate

Cryo-EM images of two proteins, apoferritin and hemoglobin, taken without and with a laser phase plate. The images are analyzed in a computer to produce detailed 3D structures of the proteins.