关键词: 多类支持向量机, 基因表达谱, 特征选取

Abstract: An approach to tumor molecular classification based on their gene expression profiles was presented. A new measure known as Between-groups to within-groups sums of squares ratio (BSS/WSS) was used as the criterion of screening predictive genes for SRBCT subtype recognition. The 152 genes were chosen by this criterion and formed the feature set whose subsets would be used to create MSVM models to identify the subtypes. The trained MSVM based on the top 25 genes ranked by BSS/WSS was able to achieve 100% accuracy on the training and blind test dataset. Then this subset was analyzed by the dissimilarity distance to remove its redundancy. As a result, the 15 genes were retained with the same accuracy as the subset of 25 genes and were regarded as the final subset. Comparison with other methods demonstrates efficiency and feasibility of the method and the predictive models proposed in this work.

Key words: Multi-category Support Vector Machine, Gene Expression Profiles, Feature Selection