Comprehensive performance evaluation of spatial transcriptomics deconvolution methods across diverse computational scenarios

Spatial transcriptome technology enables researchers to spatially reconstruct transcription patterns based on location information. However, due to the limitation of low resolution, gene expression data from spatial spots may be a mixture results from multiple cells. To uncover the cell type composition within each spot, many researcher groups have developed spatial transcriptome deconvolution methods. In this study, we conducted a comprehensive benchmarking of 19 existing methods, evaluating them with respect to accuracy, robustness and usability. We compared these methods under varying conditions, including different numbers of genes, spots, and cell subtypes. Our findings indicate that MACD, DestVI, and RCTD are the top methods for performing cellular deconvolution overall. We also summarized the experimental results and provided guidelines to assist researchers in quickly selecting the most suitable method.