# STsisal: a reference-free deconvolution pipeline for spatial transcriptomics data

## TL;DR

STsisal is a new method for analyzing spatial transcriptomics data without needing single-cell references, allowing precise identification of cell types in complex tissues.

## Contribution

STsisal introduces a reference-free deconvolution pipeline that adapts the SISAL algorithm for spatial transcriptomics data.

## Key findings

- STsisal outperforms existing methods in identifying cell types from spatial transcriptomics data.
- The method integrates marker gene selection and simplex identification to enhance precision.
- Extensive simulations and real data applications confirm the robustness of STsisal.

## Abstract

Spatial transcriptomics has emerged as an invaluable tool, helping to reveal molecular status within complex tissues. Nonetheless, these techniques have a crucial challenge: the absence of single-cell resolution, resulting in the observation of multiple cells in each spatial spot. While reference-based deconvolution methods have aimed to solve the challenge, their effectiveness is contingent upon the quality and availability of single-cell RNA (scRNA) datasets, which may not always be accessible or comprehensive. In response to these constraints, our study introduces STsisal, a reference-free deconvolution method meticulously crafted for the intricacies of spatial transcriptomics (ST) data. STsisal leverages a novel approach that integrates marker gene selection, mixing ratio decomposition, and cell type characteristic matrix analysis to discern distinct cell types with precision and efficiency within complex tissues. The main idea of our method is its adaptation of the SISAL algorithm, which expertly disentangles the ratio matrix, facilitating the identification of simplices within the ST data. STsisal offers a robust means to unveil the intricate composition of cell types in spatially resolved transcriptomic data. To verify the efficacy of STsisal, we conducted extensive simulations and applied the method to real data, comparing its performance against existing techniques. Our findings highlight the superiority of STsisal, underscoring its utility in capturing the cell composition within complex tissues.