CANCER SUBTYPE DISCOVERY AND INFORMATIVE GENE IDENTIFICATION WITH GENE EXPRESSION PROFILES
基于基因表达谱的肿瘤分型和特征基因选取

Because of their significance in cancer diagnosis and treatment, discovering new subtypes within cancers and identifying the characteristically expressed genes for each subtype based on gene ex-pression data analysis become more and more popular in recent research. A major challenge to find new subtypes is to eliminate noise, such that the newly discovered subtypes are biologically relevant and not related to experimental conditions. Several algorithms were presented in the literature but most of them gave insufficient consideration to this problem. An unsupervised filtering method to remove the "noise-genes" firstly, and then the samples are clus-tered using Bayesian Finite Mixture Model, finally the characteristically expressed genes in each subtypes are identified based on the clustering results by means of relative entropy. This process was applied to two data sets, one from an oligonucleotide gene chip experiment for leukemia samples, and the other from a cDNA microarray experiment for lymphoma samples. The discoveries have very explicit biological meanings.