Background modeling, Quality Control and Normalization for GeoMx RNA data with GeoDiff

Abstract

BackgroundNanoStrings GeoMx Digital Spatial Profiler (DSP) RNA assay can measure mRNA from hundreds of regions of customizable shape and size, yet it gives unique challenge in Quality Control(QC) and normalizating due to the omnipresent background noise incurred by the non-specific probe binding, which could not be addressed by conventional methods. Results and discussionUsing Poisson Background model, Background Score Test, Negative Binomial threshold model and Poisson threshold model for normalization from the R package GeoDiff, we perform tasks including size factor estimation, QC and normalization on GoeMx RNA assay data. They are shown to outperform conventional methods like Limit of Quantification for QC as to consistency/false positive rate and 75% quantile normalization as to eliminating technical variability and recovering true signal. ConclusionsWe present a statistical model based workflow for QC and normalizing GeoMx RNA data using GeoDiff, justified by statistical theory and validated by real/simulated data.