面向表观遗传学中组蛋白修饰的大间隔机特征选择和识别

The histone modifications in epigenetics is closely related to preventions and therapies of diseases. However, the feature numbers of histone modifications dramatically increase, which limits accurate recognitions of epigenetics and corresponding diseases. Based on urgent demands for developments of national health service, the project aims to research on the feature selection and recognition for histone modifications in epigenetics, and discloses inherent relations between histone modifications and diseases especially cancers. Firstly, transforms the DNA sequence to feture vectors which can be used by classifiers.Then,mines priori knowledges of histone modifications and introduces them into the large margin machines as regularization items,concretely, 1)establishes the large margin discriminant analysis based feature selection approach, and thus forms the uniform framework of feature selection and extraction based on large margin machines. 2)proposes the wrapper structural regularized large margin machine recursive feature selection approach. 3)presents the sparse embedded feature selection appraoch based on structural regularized large margin.4)ensembles the proposed feature selection approaches above. Ultimately, applies the proposed feature selection approaches to reocgnize low and high levels of histone modifications,and combines variants of morphologies and functions of organism to disclose the inherent relations between histone modifications and the corresponding diseases. The comprehensive studies of the projection can provide the foundations and assistances for preventions and targeting therapies of disease especially cancer as well as investigations and developments of specific new medicines.

表观遗传中组蛋白修饰与疾病特别是癌症的防治密切相关。然而,组蛋白及其修饰特征的急剧增加限制了表观遗传及相应疾病的准确识别。结合国民医疗服务发展的迫切需要，本课题旨在研究面向表观遗传学中组蛋白修饰的特征选择和识别，揭示其与疾病间的内在因果关系。首先，将DNA序列转换为分类器能够接收的特征向量。然后，挖掘组蛋白修饰数据的先验知识并引入到大间隔机中作为正则化子，具体的，1）建立基于大间隔判别分析特征选择方法，形成大间隔机的特征选择与抽取的统一框架。2）提出封装结构正则化大间隔机递归特征选择方法。3）提出稀疏结构正则化大间隔机嵌入特征选择方法。4）集成上述特征选择方法。最后，将提出的特征选择方法用于识别组蛋白修饰的水平，并结合机体的形态和功能改变，揭示组蛋白修饰与疾病发生特别是癌症间的内在因果关系。本课题研究能够为疾病特别是癌症的预防和靶向治疗及特异性新药研制提供研究基础和辅助支持。

表观遗传中组蛋白修饰与疾病特别是癌症的防治密切相关。然而,组蛋白及其修饰特征的急剧增加限制了表观遗传及相应疾病的准确识别。结合内蒙古自治区国民医疗服务发展的迫切需要，本课题旨在研究面向表观遗传学中组蛋白修饰的特征选择和识别，揭示其与疾病间的内在因果关系。首先，对组蛋白修饰数据进行分析和预处理并转换为分类器能够接收的特征向量。然后，挖掘组蛋白修饰数据的先验知识并引入到大间隔机中作为正则化子。最后，将提出的特征选择方法用于识别组蛋白修饰的水平，并结合机体的形态和功能改变，揭示组蛋白修饰与疾病发生特别是癌症间的内在因果关系。研究发现组蛋白修饰可以与基因抑制有关,也可以与基因的激活相关,这取决于被修饰的位置和程度，并且组蛋白上相同氨基酸残基不同修饰之间也会发生协同或者拮抗。全局与局部特征抽取在生物特征选择和识别中建立两个定理两个推论，对比研究显示所提出的GmLcDA算法在生物特征识别及目标分类等中取得了优秀的识别性能。大间隔机特征抽取方法aSVMDRC取得了有效的识别效果。本课题研究能够为疾病特别是癌症的预防和靶向治疗及特异性新药研制提供研究基础和辅助支持。