Study Of Traditional Chinese Medicine Formula Network And Drug Community Detection Based On Drug Properties

Using data mining techniques to explore the law of traditional Chinese medicine formula compatibility will contribute to more specific and in-depth understanding of the traditional Chinese medicine system. Common used data mining models are mostly based on transaction items including:classification-based data mining model, clustering-based data mining model and association rule mining-based model. Lots of traditional Chinese medicine theory and research are based on these, which play an important role for the exploration of the theoretical system of traditional Chinese medicine, nevertheless they are not conducive to deep mining. The recent research trend is building network models on Chinese medicine formulae data and executing complex network analysis on that basis looking forward to more deep mining. For that reason, this paper will start from perspective of complex network analysis, and put forward a new network model based on the combination of the attribute information of drugs themselves and the association information of drugs in formulae. The main work of this paper as follows:1) Data preprocessing of the original drugs and formulae data, mainly includes "the processing of drug synonyms" and "the standardization of effect terms" which lay the foundation for the later data mining algorithms.2) Extraction of the "flavor","tropism" and "effect" as drug properties, establishment of the drug attribute model in the vector space model, and definition of drug attribute similarity.3) Put forward a new measurement model to measure drug relevance, and on this basis, establish traditional Chinese medicine formula network(TCMF), and analysis the statistical characteristics of the network.4) Put forward an overlapping community detection algorithm named "SA-Fuzzy Cluster" based on the union of structure and attribute similarity to dig the drug groups with close contact.5) At last, experimental test are executed on "atrophic lung disease" formulae. The experimental results show that our algorithm is both effective and novel.