BayraktarLab/cell2location

What it solves

Cell2location addresses the challenge of mapping fine-grained cell types within spatial transcriptomics data. It solves the problem of resolving which specific cell types are present and in what abundance at various locations in a tissue, while accounting for technical noise, platform effects, and contaminating RNA that often obscure biological signals.

How it works

The project uses a principled Bayesian model to integrate single-cell RNA-seq (scRNA-seq) data with spatial transcriptomics data. It takes reference cell type signatures derived from scRNA-seq and decomposes spatially resolved RNA count matrices into these signatures. By borrowing statistical strength across different locations and modeling technical variance, it estimates the most likely combination of cell types and their abundances that would produce the observed mRNA counts.

Who it’s for

It is designed for researchers in bioinformatics and computational biology who need to create comprehensive cellular maps of diverse tissues using integrated single-cell and spatial transcriptomics.

Highlights

• High Resolution: Resolves fine-grained cell types with higher sensitivity than previous tools.
• Bayesian Framework: Uses a principled probabilistic approach to handle technical variation and unexplained variance.
• Integration: Seamlessly combines scRNA-seq reference signatures with spatial data.
• Scalable Architecture: Built using Pyro and scvi-tools, with future goals to scale to millions of locations using amortized inference.