Application of Optimization on Exploring Biomarkers and Improving the Classification Performance in Lung Cancer

• Main Authors: Chih-Hung Cheng, 鄭智鴻
• Other Authors: Austin H Chen
• Format: Others
• Language: zh-TW
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/23735356179549213772

碩士 === 慈濟大學 === 醫學資訊學系碩士班 === 102 === Lung cancer is the leading cause of death from cancer in the Taiwan and in the world. The high mortality rate (80–85% within 5 years) results, in part, from a lack of effective tools that can diagnose the disease at an early stage. The mortality could be greatly reduced through the development of sensitive molecular markers, if these biomarkers are detectable in the earliest stages. The goals of this thesis, therefore, are to identify the gene biomarkers of lung cancer and to find the best classifier by optimizing the SVM parameters. In this study, three gene selection methods: ANOVA, q-value, and Random Forests, are used to identify the potential gene biomarkers. Furthermore, we also develop three automatically optimization methods in order to have the better classification performance. These three methods are: Genetic Algorithm- Support Vector Machine (GA-SVM), Particle Swarm Optimization- Support Vector Machine (PSO-SVM), and Ant Colony Optimization- Support Vector Machine (ACO-SVM). In summary, we found twelve important genes. Among these genes, CAV1、SFTPC、EPAS1、PECAM1、EDNRB and FHL1 had been identified to have the relationship with the lung cancer; the other six genes: ZFP106、TNXB、CD36、VWF、NPR1 and LDB2 could have strong relationship with the lung cancer but still need further validation. All our three optimization methods have better classification accuracy when compared to the traditional SVM. The average accuracy of these methods can reach around 95% that is almost 6% higher than the traditional method. In these methods, ACO-SVM method is the best classifier based on our study.