论文部分内容阅读
Gene expression microarray studies have opened a window of opportunity into understanding the way in which aberrant pathways interact in tumors.We compare results from Backward Chaining Rule Induction (BCRI), a semi-supervised discovery approach designed to discover interactions in the data, with results from methods based on analysis of the enrichment of specific genes from known pathways to analyze gene expression array data from a subset of the estrogen receptor negative (ER) breast carcinomas that has been hypothesized in the literature to exhibit androgen receptor signaling (AR+).Using LefeMiner as an example of a gene set enrichment approach; we confirmed that the data supports AR signaling as a significant pathway that differentiates this subset form the remainder of ER-tumors.We applied Backward Chaining Rule Induction (BCRI) and compared the results.Again, AR signaling was identified as a pathway that differentiates this subset, validating BCRI as a strategy.In addition to AR signaling, ER response genes were also identified as significant by both LefeMiner and BCRI.We applied Geneshaving to discover genes that cluster with AR, and then analyzed the data for interactions between the AR cluster and other clusters that were associated with the ER-AR+ subtype.These included a Her2neu duster, and a cluster notable for the upregulation of EGFR trafficking genes.The relative strength of these interactions was studied using Robust Bayesian Network Analysis.