離散化基因表現資料以重建基因調控網路

Abstract

正確地預測生命體中轉錄因子與基因之間的相互關係，也就是所謂的基因調控網路，是生物資訊研究者們正熱烈探討的問題之一。微分方程式或貝氏網路常被應用於此問題的解決，但有鑑於這些方法的繁瑣或計算複雜度較高，本研究基於”轉錄因子與受調控基因之間的關係，確實存在於基因表現資料”的前提，將基因表現資料轉換成字串的形式，並利用字串排比，找出潛在的調控模組。本方法除了所需計算時間非常短之外，亦可解決基因表現中”時間差(Time Shift)”之問題。另外，對於基因表現資料並非全局相似之基因組，我們的方法亦提供預測的可能性。最後，我們以SCPD資料庫中的26個酵母菌的轉錄因子所組成之調控模組為預測對象，並將實驗數據與前人之結果作比較，發現本研究方法提升其中18個轉錄調控模組的預測能力。

To discover the relations betweens genes, e.g. genetic networks, is one of the prominent topics in bioinformatics. Bayesian networks and differential equations have been widely applied to solving the problems, and yet the complexity of these approaches also limits their performance. We propose a fast and effective method to predict transcription modules, and then to reconstruct genetic network. Our method considers not only the “time shift” issue but also the gene pairs whose expression data are partially correlated. We applied our methods to 26 known transcription modules of yeast, and compared the results with previous works based on SCPD. The results indicate that within 26 transcription modules, our method increased 18 transcription modules’ precision. In addition, we also tested our method for the capability of reconstructing genetic networks, using cell cycle-related genes and transcription factors.