Method Research On Relationship Of Gene And Disease Based On Human Phenotype Ontology

The emergence of the Next Generation Sequencing(NGS) technology accelerated the speed of the genetic data generating. Such a large amount of data make today's problem from sequencing to how to effectively use them.Phenotypes are characteristics expressed in organism, which influenced by genes, environment, etc. Thus, phenotype, gene and disease connect to each other closely. Until now, the study of phenotypic is an important link between genes and disease, and the research of phenomics has become an important mean to discover disease genes and diseases. There are mainly three kinds of methods on the study of relationships among gens, phenotypes and diseases. Firstly, mine relationships of biomedical entities from the biomedical literature using text mining and related technologies. Another method is building network on the existing relationships among phenotypes, diseases, genes, proteins and other entities, and finding new relationships. The last method is computing the similarities between entities based on ontologies which are structured knowledge system, such as gene ontology and phenotype ontology. The three methods have their advantages and disadvantages. The study of phenotype similarity computing method based on ontology can help predict patients' causative gene and disease, taking full advantage of ontology.This paper tried to study the relationship between phenotypes and genes and relationship between phenotypes and diseases based on Human Phenotype Ontology(HPO), and then, predicting patients' causative genes and diseases.Combing the HPO terms' information content with the HPO's directed acyclic graph structure, this paper proposed a similarity computing method based on paths in HPO. The experiment results showed that this method was superior to other main methods based on ontologies when predicting causative genes and diseases on different data sets-optimal data, noise data, imprecision data and noise&imprecision data. For example, this method rose 17.3% than the second best method, Resnik, when predicting causative genes in noise&imprecision data set and 18.1% than Resnik when predicting diseases in noise&imprecisiondata set.Some researches showed that diseases which belong to same class or relate to functional similar genes aggregate in disease and genes networks. Some phenotypes-noise phenotypes, which are unrelated to the patient's causative genes or diseases-unavoidably appear on the patient's body. According to the gathering feature in phenotype network, filtering the noise phenotypes from the phenotype sets can raise the accuracy of prediction. This paper built the phenotype networks, and using Page Rank algorithm on the networks to find central phenotypes and peripheral phenotypes. And then, extract the noise phenotypes from the phenotype sets to implement the noise filtering work.According to the simulation experiments, this method could effectively find the noise phenotypes, whose inverse number is 0.136. That means this method can improve accuracy rate of predicting.