论文部分内容阅读
目的研究一种高效的基因特征提取方法,以尽可能地克服传统噪声基因剔除法中阈值设置主观性带来的信息丢失问题。方法收集Golub等发布的急性白血病基因表达谱公共数据库中的数据。相对宽松地剔除噪声基因,适当增加被选基因数量,进而利用二维主元分析法(2D-PCA)技术进行二次基因特征提取,并采用基于机器支持向量机(SVM)的分类形式。结果文中方法可获得90个二次特征和100.00%的分类精度;与直接利用一次特征进行分类相比,分类精度可提高2.78~8.35%。结论通过适当增加被选基因数量提取高效且维数相对较低的特征是可行的。
Objective To study an efficient method of gene feature extraction to overcome the problem of information loss caused by the subjectivity of threshold setting in traditional noise gene knockout method. Methods The data from the public database of gene expression profiling of acute leukemia published by Golub et al. The noise gene was removed relatively loosely, the number of selected genes was increased appropriately, and then the secondary gene feature was extracted by 2D-PCA technique and the classification based on SVM was used. The results of the method can obtain 90 secondary features and 100.00% of the classification accuracy; compared with the direct use of a feature classification, classification accuracy can be increased 2.78 ~ 8.35%. Conclusion It is feasible to extract features with high efficiency and relatively low dimension by appropriately increasing the number of selected genes.