[1883] Backward Chaining Rule Induction Using Multiple Genomic Data Types To Understand Gene Interactions in Ovarian Cancer

Srinivasa C Chekuri, Mary E Edgerton. UT MD Anderson Cancer Center, Houston, TX

Background: Ovarian serous cystadenocarcinomas (OvSC) are the largest malignant tumor category of ovarian cancer. We worked to derive molecular mechanisms relevant to survivalin OvSC using multiple genomic data types as input.
Design: We used mRNA, methylation, and microRNA data for 456 OvCA samples from the cancer genome atlas project (TCGA). Patients were divided into 2/3 training (304) and 1/3 test sets (152) with similar survival distribution. An additional external validation dataset was constructed using 118 samples from a separately published research study. mRNA was normalized using Combat and Robust Multichip Analysis (RMA). A set of genes was identified using supervised principal component analysis (SPCA) to separate high and low survival groups in the training data based on Cox scores. Rank normalized hierarchical clustering (RNHC) was performed on these selected genes to generate two patient clusters. Kaplan Meier (KM) analysis was performed to evaluate significance of separation of survival for the two patient clusters in the test sets. Two genes with the highest Cox scores were selected for network analysis using Backward Chaining Rule Induction (BCRI) with mRNA, methylation status, and miRNA as inputs.
Results: KM results for separable patient clusters using 15 genes identified in the training set were significant in the TCGA test data (p=0.0033) and nearly significant in the external test data (p=0.0601). BCRI results for the top two genes based on Cox scores demonstrated a combination of methylation status and gene expression in networks generated for the high and low survival groups. Network genes consist of multiple DNA processing and G-coupled receptors including HHEX, RNF113A, DCI with methylation status for EDNRB, HIST1H2BC.
Conclusions: OvSC is a complex disease. BCRI is a methodology that allows the simultaneous evaluation of gene expression along with methylation status and miRNA expression. BCRI has potential to help us understand complex interactions of gene expression and regulation that combine to effect aggressive behavior in ovarian cancer. Genes that function to process mRNA and that effect G-coupled signaling networks appear to be important in ovarian serous adenocarcinoma.
Category: Special Category - Pan-genomic/Pan-proteomic approaches to Cancer

Tuesday, March 20, 2012 9:30 AM

Poster Session III # 269, Tuesday Morning


Close Window