Benchmarking cell-type deconvolution in cross-platform transcriptomic data

Abstract

BackgroundTranscriptomic data from diverse measurement technologies are widely used to study tissue heterogeneity. Cell-type deconvolution, which resolves mixed transcriptomic signals into cellular components, is a key analytical approach. However, achieving accurate deconvolution across platforms remains challenging due to platform-specific experimental and technological biases. ResultsWe systematically benchmarked deconvolution performance using real-world cross-platform datasets and simulated data modeling distinct technological features. Our analyses provide practical guidelines for experimental design and promote more robust and comparable cross-platform transcriptomic analyses. ConclusionLog-normal regression methods such as SpatialDecon demonstrated the most reliable and consistent performance across both simulated and experimental settings, establishing a robust framework for accurate cross-platform transcriptomic deconvolution. Moreover, our results highlight potential caveats in deconvolution predictions arising from specific experimental conditions, providing guidance for more informed experimental design and interpretation Graphical Abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=60 SRC="FIGDIR/small/690141v1_ufig1.gif" ALT="Figure 1"> View larger version (21K): org.highwire.dtl.DTLVardef@585cbdorg.highwire.dtl.DTLVardef@13088c7org.highwire.dtl.DTLVardef@163ea75org.highwire.dtl.DTLVardef@b5ccf7_HPS_FORMAT_FIGEXP M_FIG C_FIG