使用雙分群法分析基因微矩陣資料

Abstract

從基因表現資料找出有意義的基因群組,長久以來都是分析微矩陣資料的一個重要課題。由於傳統演算法在先天上的限制,許多雙分群演算法被發展出來,用以解決此問題,並有著不同的目標和策略。我們基於分析頻繁項目集的架構下,在此提出一個雙分群法。和以往較為不同的是,我們把微矩陣資料的雙分群問題,轉換為挖掘頻繁項目集的問題。為了驗證我們演算法可行,我們首先和代表性的數個傳統分群法進行比較 ,而實驗結果顯示我們的方法穩定度和精確度比傳統方法好。接著,我們也和近年來的幾個雙分群系統 ,在已知且公開的資料下,進行一連串比較。最後,將顯示我們演算法在多個測試項目下,確實超越近年來的知名雙分群法。

Finding meaningful clusters of gene expression data has always been one of the most important topics of microarray data analysis. Due to the limitations of conventional clustering algorithms, numerous biclustering methods, with different aims and strategies, have been developed to mitigate the problems. We propose a new biclustering algorithm under the framework of market basket analysis focused on frequent itemset analysis. Unlike previous works, we transform the biclustering problem into a frequent itemset finding task where significant biclusters are described as frequent itemsets. To verify its feasibility, we first compared it with several representative conventional clustering algorithms. The experiments show very promising results. We also conducted a comparative study of current biclustering systems based on the widely-used prior knowledge, Gene Ontology. The study demonstrates that our method significantly outperforms the current biclustering algorithms in our tests.