Probabilistic Modeling for Single-Cell and Spatial Transcriptomics with scvi-tools
Instructor: Can Ergen, MD
Assistant Professor, University of Würzburg; Principal Investigator, Helmholtz Munich
Overview:
Recent advances in single-cell and spatial transcriptomics have transformed our ability to characterize complex tissues. However, extracting robust biological insights from increasingly large, heterogeneous datasets requires statistical models that explicitly account for technical noise, batch effects, and allow using these large datasets as reference for novel datasets.
This workshop provides a practical introduction to probabilistic deep learning using scvi-tools, the leading open-source framework for generative modeling of single-cell and spatial omics data. Participants will learn both the theoretical foundations and practical workflows for reference mapping, cell type annotation, differential analysis, and analysis of spatial transcriptomics data using modern variational autoencoders. The workshop combines concise lectures with extensive hands-on sessions using reproducible Jupyter notebooks. Participants are expected to have basic familiarity with single-cell RNA-seq analysis and Python. Participants will leave the workshop with practical experience applying state-of-the-art probabilistic models to single-cell and spatial transcriptomics datasets. They will gain the knowledge required to incorporate scvi-tools into their own research workflows and understand how emerging AI technologies are shaping the future of computational biology.
Learning Objectives:
After completing the workshop, participants will be able to:
- Understand the principles of probabilistic generative models for single-cell data
- Train and interpret latent variable models using scVI, perform differential expression and abundance analysis
- Perform automated cell type annotation with scANVI
- Map new datasets onto existing reference atlases
- Analyze spatial transcriptomics using resolVI and scVIVA
- Understand emerging AI interfaces for biomedical data analysis, including agentic workflows and retrieval-augmented generation (RAG)
______________________________________________________
8:30 - 9:00 AM: Workshop Registration Check-in
9:00 AM - 12:00 PM: Morning Session
Introduction to single-cell RNA sequencing and probabilistic modeling
- Why probabilistic modeling matters for single-cell data
- Variational autoencoders and latent representations
- Accounting for batch effects
- Current applications and reference atlases
- Introduction to scANVI
Hands-on session | AnnData and scVI
Participants will work through an AnnData introduction and a scVI tutorial.
12:00 - 1:30 PM: Lunch Break
1:30 - 5:00 PM: Afternoon Session
1:30 - 2:30 PM: A short lecture followed by hands-on analysis
- Sample-level analysis with MrVI
- Differential expression and abundance analysis
- Assessing annotation confidence and uncertainty
2:30 - 3:00 PM: Break
3:00 - 4:00 PM: Spatial Transcriptomics
Lecture and hands on tutorial covering:
- Probabilistic modeling for spatial transcriptomics
- Addressing noise and inferring cell types in spatial data
- Inferring spatial cell states
4:15 - 5:00 PM: Outlook: AI-native analysis with scverse
Lecture exploring emerging directions in computational biology:
- Agentic interfaces for computational biology
- Natural-language interaction with scverse workflows
- Retrieval-augmented generation (RAG) for discovering reference models
- Contributing to scvi-tools: repository structure and ways to get involved
Please note, this session is in-person only.
Click the "View Session" button below to join the session.