SubID, a non-median dichotomization tool for heterogeneous populations, reveals the pan-cancer significance of INPP4B and its regulation by EVI1 in AML

Our previous studies demonstrated that INPP4B, a member of the PI3K/Akt signaling pathway, is overexpressed in a subset of AML patients and is associated with lower response to chemotherapy and shorter survival. INPP4B expression analysis in AML revealed a right skewed frequency distribution with 25% of patients expressing significantly higher levels than the majority. The 75% low/25% high cut-off revealed the prognostic power of INPP4B expression status in AML, which would not have been apparent with a standard median cut-off approach. Our identification of a clinically relevant non-median cut-off for INPP4B indicated a need for a generalizable non-median dichotomization approach to optimally study clinically relevant genes. To address this need, we developed Subgroup Identifier (SubID), a tool which examines the relationship between a continuous variable (e.g. gene expression), and a test parameter (e.g. CoxPH or Fisher's exact P values). In our study, Fisher's exact SubID was used to reveal EVI1 as a transcriptional regulator of INPP4B in AML; a finding which was validated in vitro. Next, we used CoxPH SubID to conduct a pan-cancer analysis of INPP4B's prognostic significance. Our analysis revealed that INPP4Blow is associated with shorter survival in kidney clear cell, liver hepatocellular, and bladder urothelial carcinomas. Conversely, INPP4Blow was shown to be associated with increased survival in pancreatic adenocarcinoma in three independent datasets. Overall, our study describes the development and application of a novel subgroup identification tool used to identify prognostically significant rare subgroups based upon gene expression, and for investigating the association between a gene with skewed frequency distribution and potentially important upstream and downstream genes that relate to the index gene.