论文部分内容阅读
The aim of the study was to derive gene expression signature of estrogen receptor status for Taiwanese breast cancer patients.Estrogen receptor status is by far the most important prognostic and predictive factor for breast cancer therapy,while microarray studies in the past decade have revealed distinct gene expression patterns between estrogen receptor positive and negative breast cancers.Twenty-eight sporadic breast cancer samples were snapped frozen with total RNA extracted and mRNA was hybridized by Affymetrix GeneChip(R) Human Genome U 133 plus 2.0 arrays.Genes were selected based on their absolute correlation with estrogen receptor status.Two high-dimensionality reduction approaches, unsupervised principle component and supervised partial least square regression, were applied for class prediction.For training data, the top 50 estrogen receptor status correlated genes were sufficient for both principle component and partial least square approaches with 96% of classification accuracy.Nine probes representing eight genes were selected by stepwise discriminative function and 100%, 82% and 88% of prediction accuracy was reported for training, test and selected test data with matched phenotype prevalence.Our study suggested that unsupervised principle component might be superior to supervised partial least square regression in task of dimensionality reduction,latent factor extraction and clinical phenotype prediction if entry genes were carefully selected.Proposed predictive model could discriminate breast cancers with positive and negative estrogen receptor status and was feasible for both Taiwanese and Chinese females, both with the same Han Chinese origin.