Yannick H. A. Leurs, Willem van den Hout, Andrea Gardin, Joost L. J. van Dongen, Andoni Rodriguez-Abetxuko, Nadia A. Erkamp, Jan C. M. van Hest, Francesca Grisoni, Luc Brunsveld | Nat Commun 2025 Oct 30;16:9598. doi: 10.1038/s41467-025-64617-2
Summary
Biomolecular condensates are droplets inside cells that have no membrane. They form when proteins and nucleic acids separate out of solution, a process called phase separation. Synthetic condensates let researchers study how these structures form and use them as cell-like model systems. Mapping their phase diagrams by hand, however, is slow and labour-intensive.
In this study, researchers at Eindhoven University of Technology present a self-driving lab that maps multi-dimensional condensate phase diagrams automatically. A pipetting system prepares the samples and an autonomous confocal microscope images them. An active machine learning model learns from each round of results and picks the next experiments, steering the search towards the phase boundary.
Imaging was done on a custom automated confocal system integrated by Confocal NL. It is built around the NL5+ line REscan confocal unit, combined with a large-field-of-view sCMOS camera and a Confocal NL motorised filter wheel. The whole system is controlled from Python, so it autofocuses, scans 96-well plates and records Z-stacks at 5 frames per second without anyone at the microscope.
The platform produced detailed and reproducible phase diagrams for several polypeptides and showed how chain length and salt affect phase behaviour. It also measured particle size, count and volume fraction. That adds functional information beyond a simple yes or no for phase separation.
Read the publication: https://doi.org/10.1038/s41467-025-64617-2