Cross-platform comparison of gene expression-based cancer molecular subtyping reveals discrepancies with exome capture methods.

Gene expression profiling in precision oncology is increasing with uncertain validity across platforms. In this study, we examined the application of PurIST, a molecular subtyping algorithm for pancreatic ductal adenocarcinoma (PDAC), across different platforms. We compared PurIST calls between matched samples processed by whole transcriptome and commercial exome capture RNA-seq. In parallel, we compared subtypes between matched samples processed by NanoString and whole transcriptome RNA-seq from the PANCREAS trial (NCT04683315). Between whole transcriptome and exome capture, subtype agreement was 81% with significant increase in basal-like subtype with exome capture. Differences in overall survival in patients with basal-like tumors compared to classical tumors did not reach statistical significance using exome capture (log-rank P = 0.061), whereas with whole transcriptome it was significantly shorter (log-rank P < 0.0001). Subtype agreement between whole transcriptome RNA-seq and NanoString was higher at 95%. PurIST results should be interpreted with caution when using exome capture methods.