The Summary Of Gene Expression Level Based On Primary Affymetrix Array Data

Over the past decades, hybridization-based mircroarray technology is widely used in thefields of life sciences. Mircroarray probe-leved analysis computes an expression estimate for eachgene and play a important role for downstream data analysis. However, the presence ofcross-hybridization, affects the computation accuracy of gene and isoform expression levels.Various mathematical models have been proposed to moded noise, in order to decrease theinfluence of noise.This thesis studies gene and isoform expression measurements for widely used Affymetrixtraditional3’ Genechips and GeneChip exon arrays. MM probes for low expression genes oftenobtain higher cross-hybridization than true signal in traditional3’ Genechips and this easily leadsto instable computation. So we propose PM-only multi-mgMOS which adops only PM probeintensities. Experiments on MAQC dataset and a real mouse embryo dataset have shown thatPM-only multi-mgMOS performs better than multi-mgMOS for relatively low expressed genes.As the analysis of alternative splicing gains increasing interest in recent years, occursalternative splicing for most of human genes, Affymetrix GeneChip exon arrays are largely usedto study alternative splicing in recent years. Traditional analysis methods can only compute geneexpression values and do not calculate isoform expression levels. So we design a gamma basedmodel for GeneChip eoxn arrays(GME), using the known relationships between genes、isoformsand probes. Expreiments on the MAQC exon array dataset and HNSCC dataset, shown that GMEobtains more accurate gene and isoform expression estimates compared with tradition methods.PM-mmgMOS and GME have been implemented into R packages, and have been includedin the large probabilistic model-based R package puma, which has been released by Bioconductor.Puma is freely available for biologists across the world and can be downloaded forhttp://www.bioconductor.org/.