| Research shows that alternative splicing phenomenon is commonly exists in higher eukaryotes. Many diseases are caused by the abnormality of alternative splicing, such as Parkinson's disease, Alzheimer's disease, Myotonic malnutrition and so on. Therefore, the research of transcriptome is of great interest in the field of biomedicine in recent years. The analysis of gene and isoform expression based on microarray data provides an approach to study the variation of transcriptome. Affymetrix exon and HTA2.0 arrays are commonly used in the measurement of transcriptome expression. The accuracy of transcriptome expression calculation will affect the subsequent analyses, such as finding differential gene expression and detection of alternative splicing event. Therefore, accurate estimation of transcriptome expression plays an important role in transcriptome research.Based on the previously devised Gamma model for exon array?GME? method to calculate gene and isoform expression, this thesis first proposes an improved method, i GME, to accelerate the computation and increase the accuracy of model optimization of the original GME method. The i GME method uses the known mappings between probes and isoforms, and models the conditionindependent probe effects. Second, in order to solve the problem that the increasing array number under a single condition easily slow the computation efficiency of i GME, we propose parallel computing to take full avantage of multi-core processor or cluster environment to enhance the efficiency of model optimization. Results show that the proposed parallel computing approach greatly improves the efficiency of model calulation and makes our models applicable on large-scale exon array datasets. Finally, we apply the propsed methods to the analysis of HTA2.0 array data by constructing the new mappings between probes and isoforms for this type of arrays. Results show that our models can accurately calculate the gene and isoform expression for the new HTA2.0 arrays. |
PDF Full Text Request