The theory of Artificial Intelligence (AI) has been thoroughly researched and successfully applied to the extraction of relationship between all kinds of items. Traditional Chinese Medicine(TCM) and western medicine have got their own theoretical basis and well developed systems on disease diagnosis and therapy, but some of the items in TCM are based on philosophical concepts of Ancient China, so they are difficult to be interpreted and hard to be quantitated. In this paper we focused on the relationship of the biological network of Zheng、 phenotypes and genes, intended to draw the patterns of Zheng-phenotype and phenotype-gene. We proposed two algorithms for the problems that existed in the patterns generation and presented the result using our methods. The main work of this paper contains: 一、 Pattern mining based on feature selection ·Zheng-phenotype patterns mining based on feature selection: The correlative dependence and influence of phenotypes is a big problem in the construction of Zheng-phenotype, normal feature selection algorithms cannot be used here. We proposed an improved feature selection algorithm based on Markov Blanket and used it to analysis the correlation between Zheng and phenotypes calculate the feature subset against Zheng and generate patterns of Zheng-phenotypes. ·Construction of diagnose model based on classification: Based on the patterns of Zheng-phenotype, we trained six classifiers using Bayesian network, Naive Bayesian, logistic regression, support vector machine(SVM), K-nearest neighbor(KNN) and decision tree, and presented the classification results given new patients' records. 二、Pattern mining based on text mining and inference network ·Construction of phenotype-gene patterns based on text mining: The records of Online Mendelian Inheritance in Man(OMIM) are manually maintained by experts in the field and have high reliability, we used the records in our paper to mine the relationship between phenotypes and genes. The relationship mining from OMIM were treated as the foundation of the phenotype-gene patterns. ·Mining the implied patterns using Label Propagation algorithm: The patterns mining from OMIM only cover a small part of the phenotypes and genes. For the rest of the phenotypes and genes, we proposed a Label Propagation algorithm based on the topology of protein–protein interactions(PPIs) network to generate the phenotypes-genes patterns.
修改评论