用MCMC的方法作基因選取與預測

Abstract

生物晶片的實驗能在短暫的時間內提供我們數以千計的基因資料；此時，如何從中找出重要的基因成為大家關心的問題。在2003年，Lee et al. 提出一個層級性的貝氏模型來選取基因，他們採用潛在變數來建立迴歸模型，然後用混合的貝氏先驗分配來執行基因選取的動作，MCMC中的Gibbs sampling是他們模擬參數的方法。在此篇論文中，我們修正了他們在基因選取與作預測的演算法，並且，我們也成功的把它運用在俱有遺傳性的乳癌資料上面，主要是區別在BRCA1和BRCA2二種腫瘤上的突變基因。

DNA micro-array experiments provide us thousands of genes data at once. How to identify the responsible genes is an important problem. Lee et al. (2003) propose a hierarchical Bayesian model for gene selection. They use latent variables to specialize the model as a regression setting, and then use a Bayesian mixture prior to perform the gene selection. The method they use to simulate parameters is Gibbs sampling, one kind of MCMC method. We modify their algorithm of gene selection and prediction in this paper. The method is applied successfully to hereditary breast cancer data to classify tumors with BRCA1 and BRCA2 mutations.