一個以電腦加速篩選癌症相關基因的兩段式方法

A Two-Phase Computational Method for the Speedup of Screening Cancer Related Genes

對於以分子醫學的方式來研究癌症的機轉而言，癌症相關基因的篩選具有關鍵的重要性，因此近幾年來陸續已有研究者發展一些以電腦輔助快速篩選基因的方法。在資訊技術上，若欲使用基因微陣列基因表現資料來針對癌症進行相關基因的篩選，可藉由特徵挑選的方法來做基因篩選。由於人類基因的數量實在是非常龐大，若是以生物實驗的方式一個個去篩選與特定疾病可能相關的關鍵基因，則所花費的時間、人力及費用成本就會很高。所以，我們試圖從數量龐大的人類基因中，以電腦分析癌症的基因微陣列基因表現資料，去快速地篩選出與某種癌症相關的基因。一般用電腦分析方法來解決此一問題時，除了要考慮到結果的有效性外，還需要考慮到分析所花費的時間成本。在許多實際的應用上，如果基因數量非常龐大時，就無法快速有效率地分析基因與特定疾病的相關性。遺傳演算法與最近鄰居分類法的併用是一個有效的特徵挑選資訊技術，常被用來找尋與特定疾病相關的關鍵基因。然而，高時間成本是此一基因篩選法的弱點，因此本文提出一個二段式的方式，探討如何將運算時間的成本有效地減少，同時兼顧基因篩選結果的有效性與穩定性。

Applying Bioinformatics methods has become a must in the exploration of progressions or causes of carcinogenesis, where correctly finding relevant genes of cancers is a vital component. Hence, there is a proliferation of research works related to gene screening via computational methods. Due to the large number of human genes, it is too costly, time-consuming and labor intensive currently to apply experimental procedures of wet lab to screen genes related to a particular cancer of interest. Hence, we propose to apply Bioinformatics methods to identify relevant genes of cancers at a lower cost. In terms of computational methods, cancer gene screening via Bioinformatics analysis on gene expression levels of cDNA microarrays can be achieved by feature selection. In practical applications, it is implausible to effectively identify a sufficiently small number of cancer candidate genes in a realistic time span especially if the number of genes that need to be evaluated is massive. The integration of genetic algorithm and k-nearest neighbor classifier (GA/KNN) seems to be an effective method for identifying genes related to a certain cancer of interest. One major disadvantage of the GA/KNN method is its extensive execution time; therefore this paper proposes a two-phase method to explore how to significantly reduce GA/KNN's computation time for screening key genes without reducing its effectiveness.