Research

Microscopy images are full of information, but how can we reliably turn this wealth of visual information into objective numbers? 

Image quantification is uniquely challenging in microscopy data as what constitutes relevant visual information is entirely dependent on the biological question of interest, and may often not be known a priori. In stark contrast to “classical” computer vision, bioimage analysis is both highly specialized when it comes to the question being investigated and highly diverse when it comes to the appearance of the images acquired to investigate this question, calling for the development of methods tailored to bioimages. 

To address this challenge, we develop novel computational algorithms to quantify complex visual phenotypes from microscopy data and apply these methods to biological discovery. We are an interdisciplinary team of engineers, computer scientists, and mathematicians, seeking to create innovative tools that combine sound theoretical foundations with advanced machine learning techniques. Our contributions span both general modality- and scale-agnostic image quantification approaches as well as tailored analysis pipelines developed through collaborative projects with experimental biologists worldwide. Since the establishment of the Uhlmann Group at EMBL-EBI in 2018, we have been continuously contributing to the fields of image processing, artificial intelligence (AI), and bioimage analysis, with a strong emphasis on close collaboration with our wet-lab colleagues to develop computational methods that address critical challenges in biology and biomedicine. 

One of our key research areas of interest is multimodal morphology quantification. Morphology is a key property of living systems and the primary qualitative readout that has been obtained from microscopy image data since long before the digital era. Although recent advances in computer vision, including our own contributions, have made it possible to turn approximate, qualitative observations of morphology from individual microscopy images into rich and precise quantitative data, we are still overwhelmingly observing living systems only one scale and one modality at the time. As a result, life science lacks the bigger picture of the complex reality of living systems, operating with mechanisms that span a wide range of scales. Relying on representation learning strategies, we develop novel approaches to extract semantically-rich and holistic representations of the underlying biology of complex systems, and work towards building models that are both predictive and generative to serve curiosity-driven biology research as well as translational biomedicine.