Benchmarking of spatial transcriptomics platforms across six cancer types

Spatial transcriptomics (ST) technologies are reshaping our understanding of tissue organization and cellular context in health and disease. However, technical benchmarking across platforms remains limited, particularly in formalin-fixed, paraffin-embedded (FFPE) clinical samples, which represent the most common tissue format in oncology. Here, we systematically benchmark five commercial ST platforms (Visium v1, Visium v2/CytAssist, Visium HD, Xenium, and CosMx) using matched FFPE human tumor sections from six cancer types. Uniquely, our study includes both sequencing-based and imaging-based platforms profiled on the same samples, enabling direct technical comparisons across spatial capture modalities. We evaluate platform performance across multiple dimensions, including transcript and UMI detection, spatial autocorrelation, cell-type recovery, and integration with spatial proteomics. We also quantify the impact of sampling strategies and area coverage on cell-type estimation, revealing trade-offs in spatial resolution versus tissue context. Notably, we present the first same-sample comparison of Xenium Multi-Tissue (377 genes) and Xenium Prime (5,000 genes), highlighting key differences in transcript recovery and spatial signal despite shared chemistry and imaging infrastructure. Finally, we integrate Visium spatial proteomics data with matched RNA profiles, uncovering widespread RNA–protein decoupling and spatial heterogeneity in concordance. Collectively, this work provides a harmonized dataset and technical reference for the spatial transcriptomics community, offering insight into the relative strengths, limitations, and design considerations associated with the high-throughput spatial profiling of FFPE tumors.