In the blink of an AI: Spotting millions of molecules inside us

Credit: Irina Khven, Martin Weigert, Gioele La Manno
Scientists at EPFL and TU Dresden have developed a new machine learning method called Spotiflow that can automatically process both 2D and 3D microscopy images with unprecedented speed and precision.
Our ability to study diseased and healthy human tissues and organs nowadays depends on identifying tiny molecules with cutting-edge microscopy. But a single experiment involves dozens of millions of molecules per sample, all of which must be identified from noisy images.
Understandably, this is a seriously labor-intensive task. Automating it is a solution, but automation must also ensure accurate diagnostics in order to advance medical research and treatment.
Now, a novel machine learning method called Spotiflow is transforming how researchers detect and map these molecular signals in tissues. Developed by scientists at EPFL and TU Dresden, this powerful new tool can automatically identify and precisely locate millions of transcript molecules as well as any microscopic object of interest.
Published in Nature Methods, Spotiflow represents a notable advancement for spatial biology: Unlike previous techniques that were slower, inaccurate, and required extensive manual adjustments, Spotiflow uses advanced deep learning to automatically process both 2D and 3D microscopy images with unprecedented speed and precision.
Spotiflow’s life began when Professors Gioele La Manno and Martin Weigert met as the first two fellows of EPFL's ELISIR program, which offers early-career scientists independence. The collaboration was further sponsored by the EPFL Center for Imaging.
At its core, Spotiflow uses a mathematical technique called stereographic flow regression, essentially helping computers to find molecules considering their surroundings as projected on a sphere.
“Spotiflow automates a previously labor-intensive process, delivering high accuracy and scalability for today’s large-scale biological experiments,” explains Albert Dominguez Mantes, a PhD student with EPFL’s EDCB program who led the software’s development.
Spotiflow can rapidly and accurately process large-scale data, which gives it the potential to accelerate advances in diagnostics and therapeutic development. Reflecting the team's commitment to advancing scientific research globally, Spotiflow is now publicly available as an open-source Python package and user-friendly plugin, giving researchers worldwide access to a new, cutting-edge technology.
List of contributors
- EPFL Institute of Bioengineering
- EPFL Brain Mind Institute
- EPFL Bioimaging and Optics Platform
- EPFL Swiss Institute for Experimental Cancer Research (ISREC)
- European Molecular Biology Laboratory (EMBL)
- EPFL Institute of Physics
EPFL Center for Imaging
EPFL School of Life Sciences (ELISIR program)
German Federal Ministry of Education and Research (BMBF)
Saxon State Ministry for Science, Culture and Tourism (SMWK)
Swiss National Science Foundation
Marie Skłodowska-Curie ITN ’EvoCELL’
ERC Advanced grant NeuralCellTypeEvo
National Centre of Competence in Research RNA and Disease Network
Peter and Traudl Engelhorn Stiftung
European Research Council (ERC CoG Piko)
Joachim Herz Foundation
Albert Dominguez Mantes, Antonio Herrera, Irina Khven, Anjalie Schlaeppi, Eftychia Kyriacou, Georgios Tsissios, Evangelia Skoufa, Luca Santangeli, Elena Buglakova, Emine Berna Durmus, Suliana Manley, Anna Kreshuk, Detlev Arendt, Can Aztekin, Joachim Lingner, Gioele La Manno, Martin Weigert. Spotiflow: accurate and efficient spot detection for fluorescence microscopy with deep stereographic flow regression. Nature Methods 06 June 2025. DOI: 10.1038/s41592-025-02662-x