The research, led by Lena Merten and Felix Roosen-Runge of Lund University together with Gudrun Lotze of Malmö University and Marianne Ahmad, also of Malmö University, focused on yellow pea seeds, a crop of growing importance for sustainable food manufacturing. Peas are prized as a plant-based protein source, but the way their internal structure varies between species, cultivars, developmental stages and processing treatments remains poorly understood. Because seeds are multi-component biological materials, their characterization must stretch across length scales, from the arrangement of macromolecules to the architecture of cells. The team set out to build a workflow capable of spanning that entire hierarchy in a single, coherent analysis.

At the heart of the method is scanning Small- and Wide-Angle X-ray Scattering, or SWAXS, performed at the European Synchrotron Radiation Facility in Grenoble on beamline ID13. In a scanning experiment, a finely focused X-ray beam is stepped across the sample point by point. At each position, the scattered intensity records how molecules and nanostructures are organized locally: wide-angle patterns speak to packing at near-atomic distances, while small-angle patterns reveal features on the scale of tens to hundreds of nanometers. Rastered across an entire seed section, these measurements assemble into a map in which every pixel carries a full scattering fingerprint of its neighborhood, effectively turning the seed into a mosaic of nanoscale structural signatures.

Scattering alone, however, says nothing about which chemical elements reside where. To add that dimension, the researchers paired the scattering scan with X-ray fluorescence, or XRF, a technique in which the excited sample emits element-specific radiation. By recording fluorescence signals alongside the diffraction patterns, the team could overlay maps of elemental composition onto maps of molecular organization. This multi-modal integration is what elevates the approach: regions of the seed with similar scattering behavior can be compared against their elemental makeup, exposing relationships between structure and composition that neither measurement could reveal on its own.

The real bottleneck in such experiments is not data collection but data interpretation. A scanning SWAXS run on a seed produces thousands of scattering patterns, each a complex curve of intensity versus angle, and traditional analysis would require fitting each one with a predefined structural model. That approach is slow, subjective and biased toward structures the analyst already expects. The Swedish team instead implemented a fitting-free, data-driven segmentation workflow built on machine learning. Rather than asking what each pattern should look like, the algorithms group patterns by similarity, letting distinct structural domains emerge from the data itself.

This unsupervised strategy allowed the researchers to identify and characterize heterogeneous regions within the pea seeds and to classify structurally distinct domains without relying on predefined models. The payoff is quantitative comparability: because the classification is generated by the same data-driven procedure for every sample, domains can be compared systematically across different seeds, cultivars, developmental stages or processing conditions. A seed that has been stored for months, germinated, or exposed to industrial treatment can be mapped with the same yardstick, turning what was once a descriptive picture into a measurable, reproducible analysis.

The implications reach well beyond one legume. Seed structure underpins germination, aging, storage stability and the response of seeds to treatment during food processing, and structural variation is written into the genetic setup of each species and cultivar. A method that captures molecular organization, cellular architecture and elemental composition in one pass gives plant scientists a new lens on early growth stages and gives food scientists a way to connect processing steps to structural consequences. The authors emphasize that the approach is broadly applicable to other hierarchically organized biological materials, positioning it as a versatile tool for both plant science and plant-based food science.

Technically, the study demonstrates how synchrotron facilities and modern data science have become inseparable partners. Beamline ID13 provided the brilliant, focused X-rays needed to interrogate the seed at high spatial resolution, and the resulting datasets, recorded under a documented experimental DOI, were rich enough to sustain a fully data-driven analysis. The workflow sidesteps the model-fitting stage entirely, which not only accelerates analysis but also guards against circular reasoning, since the segmentation is derived from the measurements rather than from assumptions about what the seed contains. Combined with the complementary XRF channel, the pipeline delivers a layered portrait of the seed: where the elements are, how the molecules are arranged, and which structural territories dominate the tissue.

For the growing plant-based food industry, the timing is significant. As manufacturers reformulate products around pea protein and other legume ingredients, understanding how seed microstructure governs texture, nutrition and processing behavior becomes a competitive necessity. The multi-scale, multi-modal framework described by Merten and colleagues opens the possibility of screening varieties for desirable structural traits, monitoring how storage and treatment alter internal architecture, and ultimately breeding or engineering seeds whose structure is optimized from molecule to tissue. What was once invisible, the quiet hierarchy inside a humble pea, is now legible, pixel by pixel, pattern by pattern.

Subject of Research: Multi-scale structural and compositional analysis of yellow pea seeds using scanning SWAXS, X-ray fluorescence and machine learning segmentation

Article Title: Integrating scanning X-ray scattering and fluorescence for multi-scale analysis of seed structure supported by machine learning tools

Article References: Merten, L., Lotze, G., Ahmad, M., & Roosen-Runge, F. (2026). Integrating scanning X-ray scattering and fluorescence for multi-scale analysis of seed structure supported by machine learning tools. Plant Methods, 22(1), Article 78. https://doi.org/10.1186/s13007-026-01585-8

Image Credits: AI Generated

DOI: 10.1186/s13007-026-01585-8

Keywords: X-ray scattering, SWAXS, X-ray fluorescence, machine learning, pea seeds, plant science, food science, synchrotron, data-driven analysis, seed structure, hierarchical organization, Plant Methods