Computational Cytometry Session
Chair: Bastian Höchst, München
Recent advancements in high-dimensional cytometry have revolutionised our understanding of cellular heterogeneity, particularly within complex immune landscapes. However, the exponentialincrease in data dimensionality presents significantchallenges for traditional manual gating strategies, which are prone to bias and lack scalability. To address this, novel computational frameworks integrating machine learning and automated clustering algorithms are essential to extractrobust, unbiased biological insights. This session highlights the development and application of cutting-edge bioinformatics tools designed to streamline data preprocessing, batch-effectcorrection, and single-cell phenotyping.
cyCONDOR: end-to-end solution for high-dimensional cytometry data analysis
Dr. rer. nat. Lorenzo Bonaguro
Group Leader Molecular and Translational ImmunomicsGerman Center for Neurodegenerative Diseases (DZNE), Bonn& University Hospital Bonn | Institute for Clinical Chemistry and Clinical Pharmacology
cyCONDOR provides a unified and open-source framework for the analysis of data from high-dimensional flow cytometry, spectral flow cytometry, mass cytometry (CyTOF), and proteogenomics (CITE-seq/Ab-seq) technologies. It accepts commonly used input formats (e.g. FCS files, including those exported from FlowJo) and emphasizes reproducibility, interpretability, and ease of use, making advanced cytometry analysis accessible not only to computational experts but also to wet-lab scientists with little or no prior bioinformatics experience.
Cluster diffusion reveals, visualizes, and quantifies the complex transitions and trajectories of the human B cell manifold
Marvin Weis
Marvin Weis1, 2, 3, Claudia Giesecke-Thiel2, 4, 5
1 Interfakultäres Institut für Biochemie, Eberhard Karls Universität Tübingen
2 formerly: Max-Planck-Institut für Molekulare Genetik, Berlin
3 Institut für Medizinische Virologie und Epidemiologie der Viruskrankheiten, Universitätsklinikum Tübingen
4 Klinik für Infektiologie und Intensivmedizin, Charité – Universitätsmedizin Berlin
5 Der Simulierte Mensch, Charité – Universitätsmedizin Berlin
Peripheral B cell subsets, including transitional and naïve, various activated and memory stages, as well as plasmablasts, are frequently characterized by flow cytometry to understand immunological health and disease states. However, different phenotyping approaches, panels, and analysis strategies make cross-study comparisons difficult and often fail to reflect the phenotypic B cell manifold with its stable populations and the interconnections between them.
We integrated multiple B cell phenotyping approaches into a unifying backbone staining. Dissatisfied by existing bioinformatics tools, we developed trajectory preserving unbiased and custom dimensionality reductions, a novel pseudotime algorithm capable of branching and merging trajectories, and novel differential abundance testing at quasi-single cell level to compare abundance of rare transitioning cells.
Starting with a clustering, we diffuse one-hot encoded cluster labels along a transition matrix and obtain individual cluster proximity values for each cell. Intuitively, rare cluster-transitioning cells stand out by their mixed cluster proximities. Forwarding cluster proximities into UMAP, we obtain CLUMAP, an unbiased dimensionality reduction with improved trajectory preservation. Alternatively, diffusing user-determined cluster coordinates, we obtain USERMAP, to our knowledge the first truly custom dimensionality reduction, conveniently enabling reproducible single cell plot layouts. Finally, where common pseudotime tools fail due to branching and merging trajectories, our USERPATH aligns a selection of connected clusters and shows marker trends, cell state density and differential abundance along the trajectory, even at rare transitional stages.
In our data, these tools reflect known and unknown paths of class switching and verify CD45RB acquisition as a major early milestone towards memory B cells.
Our algorithms mostly use matrix multiplications and are therefore fast, deterministic, and mathematically simple. Given their robustness, versatility, and novel capabilities, we anticipate they will make bioinformatic analyses of flow cytometry data more rational, scalable, and accessible – for B cells and beyond.