JavaScript is required to use this site. Please enable JavaScript in your browser settings.

Mapping Single-Cell Data between Species by Combined Deep Learning and Explainable AI

Title: Mapping Single-Cell Data between Species by Combined Deep Learning and Explainable AI

Project duration: 01.01.2022 – 30.06.2025

Research Area: Life Science and Medicine, Medical AI

In the project “Mapping single-cell data between species by combined deep learning and explainable AI”, we develop a principled approach to compare the cell-type- and disease-state specific molecular response between human disease conditions and corresponding animal models at the single-cell RNA sequencing (scRNAseq) transcriptome level. We apply our approach in two contexts: First, we map molecular states of whole blood samples between human COVID-19 patients and two hamster species developing moderate (Mesocricetus auratus) or severe disease course (Phodopus roborovskii) following SARS-CoV-2 infection. Second, we focus on side-effects following immunomodulatory therapies. We identify corresponding states of immune-related toxicities between cynomolgus monkeys (Macaca fascicularis) and humans in peripheral blood mononuclear cell subpopulations (PBMCs).

Overview of the first application: Cross-species scRNA-seq disease state matching between human and hamster species in COVID-19

Aims

  • transcriptome-based quantification of disease state similarities across species
  • integrating deep learning with interpretable approaches
  • establishing a respective robust, general methodologic framework
  • improving the transferability of pre-clinical animal models towards human disease conditions

Problems

  • technological and biological batch effects to be considered
  • sparse, noisy, and high-dimensional data
  • challenging biological interpretation

Technology

  • single-cell RNA-sequencing analysis
  • deep learning, variational autoencoder
  • empirical Bayes moderation

Outlook

  • Extension to other diseases and conditions for preclinical research and drug development
  • Method developments to improve explainability of findings

Publications

  • preprint: DOI: 10.1101/2024.01.11.574849v1

Team

Lead

  • Markus Scholz (IMISE)
  • Geraldine Nouailles (Charité)
  • Kristin Reiche (Fraunhofer IZI),

Team Members

  • Holger Kirsten (IMISE)
  • Martin Witzenrath (Charité)
  • Vincent David Friedrich
  • Peter Pennitz (Charité)

Partners

funded by:
Gefördert vom Bundesministerium für Bildung und Forschung.
Gefördert vom Freistaat Sachsen.