10,000 times faster than traditional methods: New computational framework automatically discovers experimental designs in microscopy

Currently, optical microscopy is the most widely used in biological sciences. The ingenuity and creativity of human researchers led to the discovery of super-resolution (SR) methods that overcome the classical diffraction limit of light around 250 nm, allowing one to resolve the organization of the smallest functional units of cellular life. Traditionally, the search for new microscopy techniques has relied on human experience, intuition, and creativity—a challenging approach given the large number of possible experimental optical configurations. For example, if an optical device consists of only 10 elements selected from 5 different components (such as mirrors, lenses or beamsplitters), there are already over 100 million unique configurations. The complexity of the field suggests that many powerful technologies may remain undiscovered, and human intuition alone may not be enough to find them. This is where AI-based exploration technology can play a huge role in exploring this space in a fast and unbiased way. “Experiments are our window into the universe, large and small scales. Given the vast number of possible experimental configurations, it’s doubtful that human researchers have discovered all the special setups. This is where artificial intelligence can help,” Explained Mario Krenn, head of MPL’s “Artificial Intelligence Scientist Laboratory”.

In order to meet this challenge, scientists from the “Artificial Intelligence Scientist Laboratory” teamed up with Leonhard Möckl, an expert in the field of super-resolution microscopy and head of the “Physical Glycoscience” research group at MPL. Together they developed XLuminA, a highly efficient open source framework with the ultimate goal of discovering new optical design principles. Researchers have taken advantage of its capabilities, focusing specifically on SR microscopy. XLuminA operates as an artificial intelligence-driven optical simulator that automatically explores the entire space of possible optical configurations. What sets XLuminA apart is its efficiency: it uses advanced computing techniques to evaluate potential designs up to 10,000 times faster than traditional computing methods. “XLuminA is the first step in bringing together artificial intelligence-assisted discovery and super-resolution microscopy. Super-resolution microscopy has provided revolutionary insights into fundamental processes in cell biology over the past few decades – with XLuminA, we Believe the story Leonhard Möckl, head of the “Physical Glycoscience” group at MPL, adds: “The pace of success will accelerate, bringing us new designs with unprecedented capabilities. “

Carla Rodríguez, first author of the work, and other members of the team demonstrated that XLuminA can independently rediscover three fundamental microscopy techniques to validate their approach. Starting from a simple optical configuration, the framework successfully rediscovers the system for image magnification. Subsequently, the researchers tackled more complex challenges, successfully rediscovering the Nobel Prize-winning STED (stimulated emission loss) microscopy and the use of optical eddy currents to achieve SR. Finally, the researchers demonstrated XLuminA’s true discovery capabilities. The researchers asked the framework to find the optimal SR design given the available optics. This framework independently discovered a way to integrate the basic physical principles of the above-mentioned SR techniques (STED microscopy and optical eddy current methods) into a previously unreported experimental blueprint. The design’s performance exceeds the capabilities of each SR technology. “When I saw the first optical design discovered by XLuminA, I knew we had successfully turned an exciting idea into a reality. Unprecedented speed, said personnel Carla Rodríguez.

The modular nature of the frame allows it to be easily adapted to different types of microscopy and imaging techniques. Going forward, the team aims to incorporate nonlinear interactions, light scattering and temporal information to be able to simulate systems such as iSCAT (interference scattering microscopy), structured illumination and positioning microscopy. The framework can be used by other research groups and customized to their needs, which will be of great advantage for interdisciplinary research collaborations.