Research On Tissue-specific Genes Based On Gaussian Mixture Model

T issue-specific genes (TS genes) are defined as genes only or mainly expressed in a specific tissue or organ, and hence responsible for specific functions and development. In order to analyze what factors make gene expresses with tissue specificity, the researchers study genes'sequence patterns, structural features and properties of regulatory factors. We account that the pattern for tissue specificity of gene expression is another factor. Studying the pattern for tissue specificity of gene expression plays an important role in understanding the biological internal mechanism, designing targeted drug and diagnosing disease. Therefore, this paper proposes the study on tissue specificity of gene expression using Gaussian mixture model. The study includes three aspects: discovering the patterns for tissue specificity of gene expression, analyzing the patterns for tissue specificity of gene expression and predicting the tissue-specific genes.The process of discovering the patterns for tissue specificity of gene expression mainly proposes the way to identify patterns, which is the clustering based on Gaussian mixture model. This algorithm is based on the probability and is more suitable for integrating the heterogeneous data. Meanwhile, the component of mixture model is more suitable to find the implied relationship between clusters and organizations.The process of analyzing the patterns for tissue specificity of gene expression mainly analyzes the experimental data using the clustering and obtains the patterns characteristic of tissue specificity patterns. The process uses the expression data from Affymetrix U133A gene chip, analyzes the clustering result compared with EST, TIGER and CFATS, evaluates the performance for tissue specificity patterns and extracts the feature. The results find some specificity patterns of gene expression for placenta, pancreatic/colon and some others.The process of predicting the tissue-specific genes mainly establishes the prediction model using pattern feature of tissue specificity and predicts the potential TS genes. The process uses the expression data from Affymetrix GNF1H gene chip and U133A + GNF1H gene chip, analyzes and evaluates the prediction performance of prediction compared with EST and TIGER. The result shows that the prediction model is good on potential TS gene prediction performance for placenta, pancreas/colon and kidney/liver, while not well enough for others.