Addressing transcriptomic assay heterogeneity for predictive modeling in cancer (RNA-seq)

The goal of this study was to evaluate how best to preprocess RNA-seq and NanoString data to, if possible, combine these data for larger cohort sizes in predictive modeling of clinical features. High-grade serous Ovarian Cancer is a highly heterogeneous disease with most data split into Microarray, NanoString, and RNA-seq. NanoString and RNA-seq showed the greatest similarities in dynamic range across these datasets. Therefore, we performed bulk RNA-seq and NanoString (PanCancer IO360 panel) on sequential samples of 26 patients to evaluate how comparable gene expression patterns were captured and if preprocessing steps could improve this. In the preprocessing steps for RNA-seq, we considered counting over genes or exons to which NanoString probes mapped.