结合基因功能分类体系Gene Ontology筛选聚类特征基因

使用两套基因表达谱数据，按各基因的表达值方差，选择表达变异基因对样本聚类，发现一般使用方差较大的前10％的基因作为特征基因，就可以较好地对疾病样本聚类。对不同的疾病，包含聚类信息的特征基因有不同的分布特点。在此基础上，结合基因功能分类体系(Gene Ontology，GO)，进一步筛选聚类的特征基因。通过检验在Gene Ontology中的每个功能类中的表达变异基因是否非随机地聚集，寻找疾病相关功能类，再根据相关功能类中的表达变异基因进行聚类分析。实验结果显示：结合基因功能体系进一步筛选表达变异基因作为聚类特征基因，可以保持或提高聚类准确性，并使得聚类结果具有明确的生物学意义。另外，发现了一些可能和淋巴瘤和白血病相关的基因。

Feature Selection for Clustering Disease Samples Based on Gene Ontology

The observation that the disease subtypes can be clustered well based on the top 10% genes expression with the highest variations across disease samples was demonstrated by analyzing two microarray datasets of both leukemia and lymphoma. It was showed that the feature genes containing strong clustering information have different distribution characteristics in the two disease datasets. Based on above observations, a new method combining gene expression profiles with gene functional knowledge to select feature genes for disease samples clustering, was proposed. After each individual gene was annotated to defined functional classes in Gene Ontology, the disease relevant functional classes enriched significantly with differentially expressed genes were identified and then the disease samples were clustered by the differentially expressed genes contained in these identified functional classes. The experimental results showed that the performance of new clustering procedure is better than that of traditional procedure. Besides, biological function comprehensions can be achieved directly with this new approach. Two feature gene sets, which may be functionally relevant to leukemia and lymphoma respectively, are extracted.