# ReSort enhances reference-based cell type deconvolution for spatial transcriptomics through regional information integration

## TL;DR

ReSort improves cell type analysis in spatial transcriptomics by integrating regional data, reducing reliance on reference datasets.

## Contribution

ReSort introduces a novel approach to cell type deconvolution by leveraging region-level data to enhance accuracy.

## Key findings

- ReSort improves reference-based deconvolution methods in simulation studies.
- ReSort reveals macrophage enrichment in a mouse breast cancer model, offering insights into immune infiltration.

## Abstract

Spatial transcriptomics (ST) captures positional gene expression within tissues but lacks single-cell resolution. Reference-based cell type deconvolution methods were developed to understand cell type distributions for ST. However, batch/platform discrepancies between references and ST impact their accuracy.

We present Region-based Cell Sorting (ReSort), which utilizes ST's region-level data to lessen reliance on reference data and alleviate these technical issues. In simulation studies, ReSort enhances reference-based deconvolution methods. Applying ReSort to a mouse breast cancer model highlights macrophages M0 and M2 enrichment in the epithelial clone, revealing insights into epithelial-mesenchymal transition and immune infiltration.

Source codes for ReSort are publicly available at (https://github.com/LiuzLab/RESORT), implemented in Python.