Comparative analysis of multiplexed in situ gene expression profiling technologies

Abstract

The burgeoning interest in in situ multiplexed gene expression profiling technologies has opened new avenues for understanding cellular behavior and interactions. In this study, we present a comparative benchmark analysis of six in situ gene expression profiling methods, including both commercially available and academically developed methods, using publicly accessible mouse brain datasets. We find that standard sensitivity metrics, such as the number of unique molecules detected per cell, are not directly comparable across datasets due to substantial differences in the incidence of off-target molecular artifacts impacting specificity. To address these challenges, we explored various potential sources of molecular artifacts, developed novel metrics to control for them, and utilized these metrics to evaluate and compare different in situ technologies. Finally, we demonstrate how molecular false positives can seriously confound spatially-aware differential expression analysis, requiring caution in the interpretation of downstream results. Our analysis provides guidance for the selection, processing, and interpretation of in situ spatial technologies.

Integrating spatial transcriptomics count data with Crescendo improves visualization and detection of spatial gene patterns

“…Accurately correcting technical effects in fluorescence in situ hybridization (FISH)-based spatial datasets is challenging because cell segmentation is a significant challenge 31,[60][61][62] .…”

Gene count normalization in single-cell imaging-based spatially resolved transcriptomics

“…This may present challenges, not only to the downstream transcriptomic analyses investigated in this study, but to other types of transcriptomic studies such as cell-cell communication analysis or analysis of impacts of cell neighborhood composition on gene expression. Additional investigation evaluating the impact of factors upstream of gene count normalization, including cell segmentation accuracy, as well as molecule detection sensitivity and specificity on in the analysis of im-SRT data is explored in Hartman and Satija [37]. Ultimately, normalization of im-SRT data should enable the removal of the effects of systematic technical variation in detected gene counts.…”

Nicheformer: a foundation model for single-cell and spatial omics

“…We first concatenate all technology-specific datasets from the SpatialCorpus-110M and pad missing genes not measured for individual cells. Previous works 52,53 have demonstrably shown technology-dependent biases between spatial and dissociated transcriptomics data. Since spatial transcriptomics measurements and the obtained signal heavily depend on the chosen preprocessing pipeline used to generate cell-by-gene count matrices 54 , average expression profiles obtained with image-based spatial transcriptomics technologies can show much higher gene counts compared to dissociated technologies 52 .…”

“…Previous works 52,53 have demonstrably shown technology-dependent biases between spatial and dissociated transcriptomics data. Since spatial transcriptomics measurements and the obtained signal heavily depend on the chosen preprocessing pipeline used to generate cell-by-gene count matrices 54 , average expression profiles obtained with image-based spatial transcriptomics technologies can show much higher gene counts compared to dissociated technologies 52 . We argue that instead of calculating a non-zero mean vector across the entire SpatialCorpus-110M, it is preferred to compute a technology-specific non-zero mean vector, which accounts better for assay-specific gene expression effects.…”