Study On Meta-Analysis Method Of Cancer Prognosis Based On Gene Expression Profiles

Many applications of microarray technology in clinical cancer studies aim at detecting organize-special molecular features for refined diagnosis. Currently, researchers developed methods to detect the prognostic features for special cancer, but the common molecular mechanism for prognosis is still unknown. This paper meta-analyzes the common prognostic signals among multiple kinds of tumors.To study the meta-analysis method, the author uses leukemia data at beginning. Rank Score has been used to preprocess microarray profile in a cross-platform way. A novel meta-analysis is designed to perform rank score (RS) on lists of genes that are derived from different studies. RS integrated with"One vs. All"(OVA) or"One vs. One"(OVO) approach of comparison is a promising way to detect signature across platform. The result is come from six published expression datasets on acute leukemia. It shows that biological signals hereafter provide stronger support of clustering than systematic differences between microarray platforms. It assists in the analysis of gene expression data for discovery, validation and treatment for subtypes of plentifully researched cancers such as leukemia.In order to discuss the existence of common prognostic mechanism for multiple kinds of cancers, this paper develops a method named Meta-Analysis Pattern Matches. The author applies this method to analyze leukemia, mesothelioma and two breast cancer data, and discovered 42 marker genes, which are the candidates of universal prognostic markers.A well-rounded method, Similarities of Ordered Gene List– SOGL, is proposed to detect the common gene expression changes between different studies. The method is complementary to previous methods. It detects significant similarities of ordered gene lists not relying on strong effects of differential gene expression in each single study, but on consistent changes across multiple studies. The ordering based approach gives top-differently expressed genes high weights, and cumulates all weighted orderings. Thus compared to previous methods, this approach can detect similarities between gene lists, even there is no significant change in one of the gene lists. After analyzing mesothelioma, glioma, prostate, and two breast cancer data including some weak effected gene list, the method can detect 5 pairs of significant similarities.A wider research on cancer prognostic markers is made on more studies, and on comparison of more gene lists. It is discovered that the ordered gene lists among prostate survival, mesothelioma recurrence, and glioma survival are similar. The responding 13 marker genes can not only improve the...