Session 3 – Data Science + AI

Thursday, 11.9.2025, 3:30 pm – 4:45 pm

Chairs: Christin Koch, Bertram Bengsch, Hyun-Dong Chang

Unveiling the power of high-dimensional cytometry data 

Modern cytometry techniques generate increasingly complex, high-dimensional datasets that challenge traditional analysis workflows. In this session, we explore how new pipelines and available tools can be leveraged to extract meaningful patterns from cytometry data and at the same time improve the accessibility of the data for improved interpretation. We discuss different strategies for data preprocessing, dimensionality reduction, and unsupervised clustering, as well as the integration of supervised models for different applications and data types (e.g. Imaging Mass Cytometry, Microbial community analysis, Immunophenotyping). Emphasis is placed on the importance of biological interpretability. Case studies will demonstrate how these workflows enhance discovery and reproducibility in single-cell analysis. Our findings highlight the transformative potential of new technologies in cytometry and the critical need for interdisciplinary collaboration between data scientists and biologists.

Christian Müller

Ludwig-Maximilians-Universität München Institut für Statistik, München

From Cytograms to Cell States: Integrating Flow Cytometry and Transcriptomics in (Bacterial) Single-Cell Analysis

Flow cytometry has long served as a cornerstone technology for profiling eukaryotic and prokaryotic cells at high throughput and single-cell resolution. While robust workflows exist for eukaryotic applications, bacterial flow cytometry presents unique challenges: overlapping populations, low signal-to-noise ratios, and rapid shifts in physiological states. Moreover, emerging technologies like bacterial single-cell RNA sequencing (scRNA-seq) provide complementary molecular insights but remain analytically siloed from cytometric data.

This talk will explore recent computational frameworks that help establish a powerful foundation for integrative next-generation single-cell analysis. In the wake of the tremendous efforts and progress achieved in human single-cell analysis, I will highlight our recents efforts in bacterial single-cell analysis, ranging from automatic bacterial flow cytometry analysis to multimodal data integration using biscot (bacterial integrative single-cell optimal transport). biscot introduces a Gaussian Mixture Model-based Optimal Transport (GMM-OT) framework that enables robust tracking of dynamic bacterial subpopulations and integrates unpaired multi-omics data. Using time-series flow cytometry of Bacillus subtilis, biscot captures subtle phenotypic transitions, such as spore germination, and projects transcriptomic states onto cytometric populations, revealing gene-level markers like spoVID and nin. The latter analysis is possible thanks to our BacSC workflow, a fully automated statistical pipeline designed specifically for the preprocessing, dimensionality reduction, clustering, and differential expression analysis of bacterial scRNA-seq data.

Together, these frameworks enable a seamless integration of phenotype and gene expression, opening the door to more nuanced investigations of microbial physiology, antibiotic responses, and microbial community dynamics at single-cell resolution. This talk will first provide a broad overview accessible to cytometrists across domains and then dive into the computational workflows tailored to bacterial systems. By bridging cytometry with transcriptomics, we propose a unified vision for bacterial single-cell analysis that mirrors the integrative power recently achieved in eukaryotic systems.

Maximilian Kloppe

Maximilian Kloppe1, Stefan J. Maurer2, Tobias Abele2, Kerstin Göpfrich2, Sebastian Aland1,3,4
1Institute of Numerical Mathematics and Optimization, TU Bergakademie Freiberg, Germany
2 Center for Molecular Biology of Heidelberg University (ZMBH), Germany
3 Faculty of Computer Science/Mathematics, HTW Dresden, Germany
4 Center for Systems Biology Dresden, Germany

High-Throughput Mechanical Characterization of Giant Unilamellar Vesicles by Real-Time Deformability Cytometry

Real-time deformability cytometry (RT-DC) enables high-throughput, contact-free mechanical characterization of giant unilamellar vesicles (GUVs).  However, the interpretation of vesicle deformation under flow has been hindered by the absence of a suitable theoretical or computational framework. Here, we present a simulation-based model that describes GUV deformation in RT-DC, taking into account the surface dilational elastic modulus as the dominant mechanical parameter. Using phase-field simulations over a wide parameter space, we find GUV deformation to depend linearly on GUV area.

Based on these results, we derive two complementary fitting strategies for extracting the surface dilational modulus K from RT-DC data: a direct model-based fit for single-vesicle characterization and a noise-resistant collective approach that enables robust population-level estimates. Furthermore, we introduce a combined fitting method that integrates both approaches to filter outliers and improve accuracy in heterogeneous or noisy datasets. All methods scale across varying flow rates, channel geometries and buffer viscosities, and produce predictions of K consistent with literature values for different lipid compositions.

Compared to traditional techniques such as micropipette aspiration, our approach offers orders of magnitude higher throughput without mechanical contact, making it particularly suitable for GUV population studies. Beyond mechanical phenotyping, this framework opens new avenues for sorting vesicle populations based on membrane mechanics, a capability of growing interest in synthetic biology and soft matter research.

Literature:
[1]  M. Kloppe et al. “High-Throughput Mechanical Characterization of Giant Unilamellar Vesicles by Real-Time Deformability Cytometry.” (2025). arXiv:2505.15341

Felix Röttele

F. Röttele1, B. Hockenjos1, L. S. Mayer1, P. Hasselblatt1, B. Bengsch1,
1 Department of Medicine II, Medical Center – University of Freiburg, Faculty of Medicine, University of Freiburg, Germany

Interactive and Explorative Tool for Imaging Mass Cytometry Data Analysis

Background & aims: Imaging Mass Cytometry (IMC) offers unprecedented spatial resolution and multiplexing for tissue analysis, but its data analysis pipeline remains challenging, especially for beginners. Pre-processing steps carry a significant risk of degrading data quality when applied without sufficient feedback or visual control. Moreover, effective, user-friendly tools for interactive pre-processing and high-resolution phenotype visualization are lacking.
To address this gap, we developed an interactive and modular Python-based pipeline that builds upon established libraries (scanpy, scikit-image, mesmer, plotly) to facilitate both pixel- and single-cell-level analyses, with an emphasis on usability, interactivity, and spatial context awareness.

Methods: The pipeline executes in four core stages:

  1. Pixel-Level Analysis:
    1. Per-channel and per-image pixel statistics (min, max, median, Otsu thresholding)
    1. Outlier and extreme-value detection for quality control
  2. Interactive Preprocessing via Napari Plugin:
    1. Grid-based multi-image visualization filtered by pixel/cell data
    1. Preprocessing operations (percentile clipping, filtering, thresholding using scikit-image)
    1. Single-cell-type segmentation with Mesmer and dynamic mask sub setting
    1. Batch processing and cell-type specific visualization at high resolution
  3. Single-Cell Data Preparation:
    1. Spatial-gating: distance-based Zonation
    1. Phenotyping via supervised Leiden clustering
    1. Block-diagonal graph-based UMAP for structure-preserving dimensionality reduction
  4. Interactive Single-Cell Visualization with Plotly dashboard:
    1. Linked scatterplots and Napari subsets for exploratory gating
    1. Boxplots with built-in t-tests for statistical comparisons