以貝氏模式利用條件自迴歸分布尋找甲基化誘導之變異基因

Identification of Methylation-driven Genes with Bayesian Conditional Autoregressive Model

鄭璽容(Shi-Jung Cheng)

指導教授 ： 蕭朱杏

共同指導教授 ： 盧子彬(Tzu-Pin Lu)

國立臺灣大學/公共衛生學院/流行病學與預防醫學研究所/碩士(2015年)

https://doi.org/10.6342/NTU.2015.02942

摘要

近年來許多相關性研究(association studies)多專注在多基因平台資料(multi-platform genetic data)的整合式分析(integrative analysis)，此類型的研究除了可以包含不同平台資料(如DNA與gene expression)代表的不同生物意義，還可以避免只利用單一平台進行分析的一些缺點，如遺傳力的解釋不佳、模型包含的資訊不足、以及研究成果難以重現等。除此之外，多平台資料的分析還可以讓科學家有機會探討不同平台的基因標記彼此之間互動的情形。
再者，單一平台的基因資料通常都是高維度數據，因此在統計分析上多為單一標記基因的檢定(single-marker test)，這類方法不但忽略同一平台內基因之間的交互作用，也可能面臨多重檢定(multiple tests)所導致的檢力的不足。因此，有些學者發展出同時考慮一組基因的方法，例如以基因集合為主的分析(gene set-based analysis)或以生物路徑(pathway analysis)為主的分析，以降低資料的維度。使用生物路徑的好處是，如此可以瞭解哪些基因參與了特定的細胞功能並且如何相互影響。換句話說，藉由生物路徑，我們可以建立基因之間的交互作用同時保留生物上的解釋意義。
本論文為了在相關性研究中考慮基因之間的關係，並且不侷限於單平台資料，提出了一個貝氏模式，並且以條件自迴歸分佈(conditional autoregressive model)來處理基因間在生物路徑中的關係。這個自迴歸分佈能同時整合受 DNA甲基化影響的基因、與RNA表現的微陣列基因資料，進而偵測對疾病狀態有影響的基因。最後，我們利用卵巢癌的存活資料來示範這個統計模式。實際資料分析的結果顯示，這個模型可以偵測在一個生物路徑中對疾病的存活有影響的基因，其中有些基因與疾病的相關性已經被其他研究學者報導過，其他基因則可能成為未來其他生物實驗室研究的候選基因。

關鍵字

基因表現 ； DNA甲基化 ； 生物路徑 ； 微陣列

並列摘要

Multiple-platform analysis has recently become the focus of many genomic research projects. Such analysis offers an opportunity to account for the interaction between genetic observations from different platforms. Additionally, it may avoid the problems encountered in the analysis with single platform genetic markers, such as low heritability, limited information and failure in reproducing findings.
Another problem faced in association studies is the fact that genetic data are often high-dimensional, and thus the most common approaches are single-marker tests. These tests cannot consider gene-gene interaction, and can lead to low statistical power due to corrections for multiple tests. An alternative is to consider sets of genes such as gene set-based analysis or pathway analysis. Through pathways, the knowledge as which genes participate in certain functions and how these genes interact with each other can then be used to construct the relations between genes in statistical analysis, while reserving the biological meaning at the same time.
In this thesis, we propose a Bayesian model with a conditional autoregressive distribution to address the relations among genes in a given pathway. This model also integrates DNA methylation and RNA expression microarray data to detect influential genes. We next illustrate this Bayesian model with an ovarian cancer study. Several influential genes are identified, where some of them have been reported earlier. Finally, we discuss issues and applicability of this proposed model for genetic association studies.

並列關鍵字

gene expression ； DNA methylation ； pathway ； microarray

參考文獻

Baylin, S.B. and Ohm, J.E. (2006) Epigenetic gene silencing in cancer - a mechanism for early oncogenic pathway addiction. Nat. Rev. Cancer, 6: 107–116.

Berns, EMJJ. et al. (1996) TP53, and MYC gene alterations independently predict poor prognosis in breast cancer patients. Genes Chromosomes Cancer, 16: 170–179.

Borczuk, A.C. et al. (2004) Molecular signatures in biopsy specimens of lung cancer. Am J Respir Crit Care Med, 170: 167–174.

Broggini, M. et al. (2000) Cell cycle-related phosphatases CDC25A and B expression correlates with survival in ovarian cancer patients. Anticancer Res, 20: 4835–4840.

Chen, Z.J. and Pikaard, C.S. (1997) Epigenetic silencing of RNA polymerase I transcription: a role for DNA methylation and histone modification in nucleolar dominance. Genes Dev, 11: 2124–2136.