Our research group uses computational analysis and machine learning algorithms to identify patterns and relationships from complex, multimodal, and multiscale datasets of the brain in health and disease. We collaborate with national and international partners to apply novel methods from the fields of computational biology and computer vision to obtain new insights from brain datasets, with a primary focus on neurovascular and neurodegenerative phenotypes. We focus on learning interpretable representations from complex high-dimensional datasets that allow us to readily explain the biological relevance of results and ultimately understand underlying disease mechanisms.

We are particularly interested in the following areas:

1) Spatial Omics:

We aim to gain a deeper understanding of the cellular organisation of the brain by analysing gene and protein expression in a spatial context. Using spatial omics datasets, including 10x Visium and MERFISH, we extract spatial gene and protein expression patterns and relate them to morphology and single-cell RNA sequencing data. For this, we have recently developed the spatial omics analysis toolbox squidpy (Palla, Spitzer, et al, Nature Methods 2021) and the subcellular analysis method campa (Spitzer, Berry et al, Nature Methods 2023) in collaboration with Fabian Theis at Helmholtz Munich.

2) Image Representation Learning:

We establish pipelines for extracting interpretable features from histology and neuroimaging data using advanced deep learning algorithms. As part of the MELD project in collaboration with Sophie Adler and Konrad Wagstyl at University College London, we are currently using graph neural networks to identify focal cortical dysplasia lesions in a large cohort of epilepsy patients (Spitzer, Ripart et al, Brain 2022 and Spitzer, Ripart et al, MICCAI 2023). Our goal is to develop and apply advanced image analysis techniques that will help us to better understand structural brain changes associated with neurodegenerative diseases.

3) Cross Modal and Cross Dataset Learning:

We aim to combine datasets across diseases and modalities into a multimodal and multiscale brain atlas. By integrating data from different sources and diseases, we hope to identify features that are conserved across diseases and study correlations between modalities. Towards this aim, together with collaborators from Forschungszentrum Jülich, we are currently investigating machine learning methods to connect spatial omics and histology in one multimodal respresentation in the context of a Helmholtz Imaging Project. 

Please contact hannah.spitzer@med.uni-muenchen.de to learn more.