Power analysis for spatial omics

Abstract

As spatially-resolved multiplex profiling of RNA and proteins becomes more prominent, it is increasingly important to understand the statistical power available to test specific hypotheses when designing and interpreting such experiments. Ideally, it would be possible to create an oracle that predicts sampling requirements for generalized spatial experiments. However, the unknown number of relevant spatial features and the complexity of spatial data analysis makes this challenging. Here, we enumerate multiple parameters of interest that should be considered in the design of a properly powered spatial omics. We introduce a method for tunable in silico tissue generation, and use it with spatial profiling datasets to construct an exploratory computational framework for single cell spatial power analysis. Finally, we demonstrate that our framework can be applied across diverse spatial data modalities and tissues of interest.

Sample size calculation for a NanoString GeoMx spatial transcriptomics experiment to study predictors of fibrosis progression in non-alcoholic fatty liver disease

“…The number of ROIs per patient affects the ability to accurately estimate the random effects (in this case, patient-specific intercept terms) and not the fixed effect regression coefficients [56]. This represents sharp contrast to the current literature that focuses primarily on spatial variability of expression within the tissue and, accordingly, increasing the number of spots or ROIs per tissue sample [5][6][7]. Whilst group differences between patients could be addressed with conventional experimental methods such as RNA-sequencing or even qPCR on FACS sorted or laser micro dissected tissue, this presents an impractical barrier in human research when spatial transcriptomics with the NanoString platform enables reliable quantification of gene expression levels from small areas of archived formalin fixed and paraffin embedded tissue.…”

“…of spots or ROIs per tissue sample [5][6][7]. Whilst group differences between patients could be addressed with conventional experimental methods such as RNA-sequencing or even qPCR on FACS sorted or laser micro dissected tissue, this presents an impractical barrier in human research when spatial transcriptomics with the NanoString platform enables reliable quantification of gene expression levels from small areas of archived formalin fixed and paraffin embedded tissue.…”

“…Yet, limited methodological research is available on this topic, especially in clinical specimens. Previous studies reporting sample size calculation for spatial transcriptomics focused on detection or co-localization of specific cell populations on the slide or identification of spatially variable features [5][6][7]. Specifically, (which was not certified by peer review) is the author/funder.…”

“…Yet, limited methodological research is available on this topic, especially in clinical specimens. Previous studies reporting sample size calculation for spatial transcriptomics focused on detection or co-localization of specific cell populations on the slide or identification of spatially variable features [5][6][7]. Specifically, despite the widespread use of the technology in unpowered exploratory research (https://nanostring.com/resources/publications/), only a handful of studies with NanoString GeoMx report dedicated sample size calculation based on pilot experiments but do not outline the requirements and statistical considerations in detail [8,9].…”

Explainable multiview framework for dissecting spatial relationships from highly multiplexed data

“…We first assess the ability of MISTy to recover purely structural relationships decoupled from the influence of functional relationships. To this end, we generated three in silico tissues with specified spatial interactions between four cell types [36]. The number of cells belonging to each of the cell types is approximately equal, to remove the potential confounding effect of abundance.…”