You are here:

Spatially aware approach for cell-type deconvolution in spatial transcriptomics data

Spatial transcriptomics (ST) [2] and single-cell RNA-sequencing offer insights into the cell type topography in a tissue. Spots contain multiple cells, therefore the observed signal conveys information about mixtures of cells of different types, which allows researchers to create cell-type mapping models. One way of inference cell types decomposition is probabilistic graphical models are used to represent both hidden and visible variables in the model. Probabilistic graphical models (PGMs) are a rich framework for encoding probability distributions over complex domains: joint distributions over numbers of random variables that interact with each other. These representations sit at the intersection of statistics and computer science, relying on concepts from probability theory, graphs, and machine learning. The graphical model approach is added in the Celloscope [1] model. The Celloscope is a Bayesian model that estimates the abundance of cell types at each location by decomposing the spatial expression count matrix into a predefined set of reference cell-type signatures. The model takes an untranslated spatial expression count matrix of genes at localization and marker genes identification as input. In this presentation, I will talk about the modeling of cell decomposition problems based on spatial transcriptomics and explain how they are modeled using probabilistic graphical models, in particular the Celloscope model and its simplification for my work. In my work, I consider the spatial interaction between the presence of types that collocate. I will talk about this idea introduced in my extension of Celloscope model assumptions.


  • [1] Agnieszka Geras et al. “Celloscope: a probabilistic model for marker-gene-driven cell type deconvolution in spatial transcriptomics data”. In: bioRxiv (2022). doi: 10.1101/2022.05.24.493193. eprint: https:/ /www.biorxiv . org/content/early/2022/05/25/2022.05. 24. 493193.full.pdf 
    https ://
  • [2] Vivien Marx. “Method of the Year: spatially resolved transcriptomics”. eng. In: Nature methods 18.1 (2021), pp. 9–14. issn: 1548-7091.

Marcin Wierzbiński is a Scientific Programmer at SANO, Center for Computational Medicine. He graduated bachelor’s degree in 2020 from the Faculty of Mathematics, Informatics, and Mechanics, University of Warsaw (MIMUW) with a related bachelor’s thesis: ”Solving SAT problems using quantum algorithms”. He gathered experience as a software engineer and research associate intern at the Polish Academy of Science – Centre for Theoretical Physics with experience in deep learning and quantum information. Currently, in his final year of mathematics with a specialization in Machine Learning at MIMUW. His master’s thesis is on ”Spatially aware approach for cell-type deconvolution in spatial transcriptomics data”. During his graduate studies, he spent a semester at the University of Oslo.

Marcin Wierzbiński
Scientific Programmer at SANO, Center for Computational Medicine

Monday, 20th June 2022, 2:00-3:30 PM (CEST)

Join via ZOOM on