Sebastian Birk

I build machine learning methods that read the spatial organisation of human tissue — where cells sit, which neighbourhoods they form, and what those neighbourhoods do.

I work in the Lotfollahi Lab at the Wellcome Sanger Institute on generative models and graph learning for spatial and single-cell genomics. My doctoral work, with the Theis Lab at the Technical University of Munich, produced NicheCompass, a method for characterising cell niches that keeps its learned representations interpretable rather than trading interpretability away for accuracy.

More recently I have been working on foundation models for tissue — TERRA, discrete tokenization of cells and niches, and generative reconstruction of tissue from expression alone.