Identification of Single Nucleotide Polymorphism Interactions Associated with Survival and Risk In Multiple Myeloma Using Novel Data Mining Methods

Abstract 2973

Disease risk and therapeutic outcomes are impacted by both tumor heterogeneity as well as germline variations found in the population. Multiple myeloma (MM) shows significant heterogeneity in genetic aberrations in tumor cells, that together with inherited polymorphisms, affects disease risk and therapeutic response. In order to identify the impact of genetic variations (SNPs) on MM we have developed a Bank On A Cure platform for examining 3404 SNPs, selected in 983 genes associated with pathways affecting cellular functions important in cancer. Using SNP data sets we sought to identify genetic interactions, beyond single univariate association analysis. The challenge was to use data mining methods that take into account relatively small cohorts of patients, in which false discovery rates typically exceed the power of the study. We report results from using novel computational approaches that efficiently identify higher order SNP interactions associated with disease risk as well as survival outcomes, while minimizing the false discovery rate. The BOAC SNP panel was used to develop a data base on 143 patients selected for short (<1yr) versus long (>3yr) survival in ECOG 9486 and SWOG 9321; as well as 247 newly diagnosed patients and equal number of controls for disease risk analysis. One algorithm developed employs a discriminative pattern mining approach in which defined pathway sets of SNPs are used in combination testing. A second algorithm used identified SNPs that had some association with outcome (survival or disease status); but demonstrated a significant increase in associations when examined in combinations – we refer to this as a p-value jump association. Variations in genes associated with cell cycle, apoptosis, drug metabolism, stress response and immunity reached very low p-values, and survived multiple comparison testing when analyzed in combinations associated with both survival (PFS) predictions as well as analysis of case-control disease risk. Some of the key genetic variations identified in various combinations, included: PTRB, PTEN, CDK5, XRCC4, GSTA4, GPX, DYPD, PCNA, CYP4F2, VEGF, PON1, ALK, and BAG3. The data mining methods and algorithms used, and specific combinations associated with risk and survival, will be presented. These results are being further validated in new cohorts, and functional implications of identified genetic variants are being investigated in HapMap cell lines.