Oligo基因芯片的异常值处理对有监督疾病分类的影响  

Effects of Outlier Treatments on Disease Classification for Oligo Microarray Data

作　　者：吕莹丽[1] 王栋[1] 郭政[1] 于梁梁[1] 李彦辉[1] 朱晶[1] 王晨光[1] 

机构地区：[1]哈尔滨医科大学生物信息学系,哈尔滨150086

出　　处：《中国生物医学工程学报》2008年第1期69-75,共7页Chinese Journal of Biomedical Engineering

基　　金：国家自然科学基金(39970397,30170515,30370388,)资助项目;国家863计划(2003AA2Z2051;2002AA2Z2052)资助项目;哈医大211工程“十五”建设项目

摘　　要：基因芯片实验产生的表达谱数据中存在大量不合格的检测点,对异常值的不同处理,对于有监督疾病分类结果的影响很大。针对此问题,在Oligo芯片数据中,在表达水平层面,通常对检测值做最大值和最小值的预处理后,进行后续分析。本研究选取了四套Oligo芯片数据集,采用不同限定芯片数据中最大值和最小值的方法,考察支持向量机、K近邻、决策树三种分类器对分类疾病样本效能的影响程度。结果显示:Dudoit等限定最大值和最小值分别为16000和100是一种合理的策略,可以达到很好的分类效果。同时发现对于小于100的检测值较多的数据集,采用限定最小值为10的策略同样能得到很好的分类效果,并可以为后续分析保留更多的原始数据。因此,合理限制Oligo芯片中的异常值,对于提高疾病分型是一种较好的策略。进一步采用功能表达谱方法,构造反映功能结点中全部注释基因的总体表达状态的均值或中值指标,利用构建的功能表达谱进行分类分析。发现不同异常值的限定方法对基于功能表达谱进行分类得到的准确率的影响较小,可以获得较稳定的分类结果。Microarray datasets frequently contain a large number of unreliable values. Different methods of restricting outliers have effect on the result of disease classification applying into supervised learning. In order to solve this problem, oligonucleotide microarray datasets need to be processed before subsequent data mining on expression level measurements. This paper focused on the effect of different outlier treatments for oligonucleotide microarray data on disease classification analysis evaluated by three kinds of classifiers, including support vector machine （SVM）, K- nearest neighbor （KNN） and decision tree （DT）. By analyzing four datasets, we demonstrated that replacing the maximum and minimum by 16000 and 100 suggested by Dudoit could get satisfying classification results. Alternately, we found that replacing the minimum by 10 was also a good choice. In addition, we employed the functional expression profiles （FEP） composed of summary measures （ the arithmetic mean or the median） of the gene expression values in each functional module to perform the classification analysis. The results show that different outlier treatments have little effect on the disease classification results performed by FEP.

关 键 词：基因表达谱 异常值分类 功能表达谱 

分 类 号：R318.08[医药卫生—生物医学工程]