Study of Bio-inspired Algorithms Applied to Human cDNA Microarray Dataset of Ovarian Cancer Target Genes

Abstract

The incidence of ovarian cancer in gynecological cancer is not high, but the mortality rate is first. Currently used as a tumor marker CA-125 screening tool, but the lack of CA-125 accuracy. Therefore, to find a better screening tool for ovarian cancer is a very important issue. This study is mainly composed of microarray dataset in ovarian cancer patients, by data mining methods, establish a distinguished model of stages of ovarian cancer, and to identify the target genes. First, the ovarian cancer microarray dataset used in filter technology to pick out important genes, then important genes used in wrapper technology to pick out target genes. The main technology wrapper is the use of Bio-inspired Algorithms, including Back Propagation Artificial Neural Network, Genetic Algorithm, Particle Swarm Optimization, Artificial Fish Swarm Algorithm. The results of this study, this study found that the best use of stages of ovarian cancer classifier, whenever it detects seven target genes ITGB2, PLEC, POSTN, C1S, FN1, CDK5, PAPPA, you can distinguish between the four of stages of ovarian cancer, accuracy up to 97.16%.The results can be used as biological and medical important reference, after proven to be the future of biological experiments and medical experiments, expectations as ovarian cancer screening tool, allowing doctors, early detection and early treatment to improve survival in patients with ovarian cancer.

卵巢癌在婦科癌症中的發生率雖然不高，但死亡率卻是第一位，目前採用腫瘤指標CA-125作為篩檢的工具，但CA-125準確性不足。因此，尋找更好的卵巢癌篩檢工具是非常重要的問題。本研究旨在由卵巢癌患者的微陣列資料，藉由資料探勘的方法，建立一套可分辨卵巢癌期別的模型，並找出標靶基因。首先使用過濾器技術從卵巢癌微陣列資料中，挑選出重要基因，再使用包裝器技術從重要基因中，挑選出標靶基因。其中包裝器的主要技術是採用仿生演算法，包括倒傳遞類神經網路、基因演算法、粒子群演算法、人工魚群演算法。本研究結果，發現利用本研究的最佳卵巢癌期別分類器，只要檢測7個標靶基因ITGB2、PLEC、POSTN、C1S、FN1、CDK5、PAPPA，即可分辨卵巢癌四個期別，正確率可達到97.16%。研究結果可作為生物學及醫學上重要的參考依據，待未來經過生物實驗及醫學實驗的驗證後，期望可以作為卵巢癌篩檢的工具，使醫生可以早期發現早期治療，提高卵巢癌患者的存活率。