Fig. 2.

Summary of benchmarking results for the 14 deconvolution methods. (a) The names of two types of methods (i.e. enrichment-based methods and deconvolution-based methods) are respectively ordered by their performances. (b) The overall ranking of the 14 methods. (c) Accuracy of different methods across three metrics and three synthetic datasets. (d) Stability of each deconvolution method with respect to sequencing depth, spot size and normalization. To evaluate the robustness of different methods to sequencing depth, for each method, we calculated the variance of the aggregated score at different sequencing depths (Supplementary Text S6). ‘Not applicable’ indicates that certain deconvolution method does not support performing normalization of ST data. (e) Usability assessment in terms of running time and memory