Fig. 1: Study overview.

Experimental setup of joint analysis of breast cancer sections using spatial transcriptomics and our computational pathology approach. Computational tissue annotation was performed on matched H&E images from Visium spatial transcriptomics analysis supported by the QuPath (version 0.5.1) platform. After stain vector correction, cell types were annotated by an object classifier using the random trees algorithm. In parallel, data from single-cell transcriptomics were downloaded and used as a reference to deconvolute spatial transcriptomics data through seven different methods. Spearman’s correlation coefficient was used to measure the performance of deconvolution methods. Subsequently, gene expression clustering and other downstream analyses were carried out using the outperformed deconvolution method, assisted by the computational tissue annotation in different ways. Created in BioRender. Li, T. (2025) https://BioRender.com/n2mhmke.