Differential expression analysis in single cell and spatial RNASeq without model assumptions

Gene up(down)regulation findings in single cell and spatial RNASeq can be inconsistent despite remarkable progress in technology. False findings in high-quality samples raise concerns about assumptions behind widely accepted data analysis approaches. We have developed a weighted averaging approach for data analysis without assuming anything besides randomness of technical noise. This approach is closely related to prior work on statistics of cluster-randomized experiments. We show that weighing transcript counts based on measured noise variances and utilizing weighted rather than standard unweighted tests reduces both false positive and false negative findings. Our approach eliminates the need for parametrizing data distributions and/or rescaling transcript counts, which may cause artifacts by distorting and biasing the data. The resulting analysis is less complex and produces more consistent differential gene expression. The present dataset is a Visium HD slide with 3 sections from the same E18 mouse tibia. Comparison of proliferating chondrocytes in the growth plates of these sections based on commonly used differential expression analysis methods produced numerous false positive findings, initiating our study. Our weighted averaging appproach eliminated these false positive findings. This dataset is used as an example in the paper for illustrating application of the weighted averaging approach to high-resolution spatial transcriptomics. Overall design: Tibias from untreated, wild type C57BL/6J mouse embryo at embryonic day E18.5 were dissected, fixed in freshly prepared 4% methanol-free formaldehyde solution in phosphate-buffered saline for 24 h at room temperature, and embedded into paraffin. Three sections from the same tibia were utilized for the experiment.