中文摘要：

生物信息学的一个重要目的是帮助人类深入地认识疾病的过程、遗传特性和潜在的治疗方法。然而,发现致病基因往往是一项复杂而艰巨的工作,比如一些遗传性的眼部疾病。在综合了收集到的众多基因表达数据的基础上,提出一种双层的直推式机器学习（TTP）模型,用于发现潜在的视网膜致病基因。里层用于从多维的Human Body Map 2.0和眼部组织基因表达谱中分别获取贡献度;在外层学习中,里层获取的贡献度将和Crx和Ch IP-Seq数据一起学习得出致病基因的排序结果。实验结果表明,在致病基因预测上,直推式学习的准确度要优于传统的监督学习。另外,还发现一个有趣的现象,数据的集成并不是总能得到有利的结果。

英文摘要：

One of the major goals of biological science is to help people understand disease process,heritability and potential treatment in depth. However,it is usually a daunting job to discover the pathogenic genes,such as some inherited ocular diseases. On the basis of colligating numerous collected gene expression data,we presented a two-layer transductive machine learning（ TTP） model used for finding potential retinal pathogenic genes. Its inner layer is in charge of gaining contribution degrees from multiple-dimensional features profile of Human Body Map 2. 0 and ocular tissues gene spectrum separately. In outer layer learning,the contribution degree obtained by inner layer will learn together with Crx and Ch IP-Seq data to derive the prioritisation of the pathogenic genes. Experimental results showed that the transductive learning method did perform better than the traditional supervised learning method in accuracy on predicting pathogenic genes. In addition,an interesting finding was that the data integration was not always helpful.