Computational methods for transcription analysis using oligonucleotide microarrays

With the rapid growth in genomic information in recent years comes the computational challenge of extracting biological insight from this wealth of data. This thesis focuses on computational techniques for analysis of one particular source of high-throughput biological information, oligonucleotide microarrays. Oligonucleotide microarrays are an emerging technology which allow the simultaneous measurement of the expression levels of thousands of gene transcripts. Traditionally, microarrays have been used to assay only the translated regions of genes. In this thesis, we present an algorithm for the analysis of oligonucleotide microarray data which identifies transcripts throughout an entire genome, including intergenic regions. Our results demonstrate that this is an effective approach for identifying untranslated transcripts on a genome-wide scale. The application of our algorithm to oligonucleotide microarray data from the microorganism Escherichia coli enabled us to identify a number of new transcripts, including previously undiscovered small RNA genes. We then improve our transcript identification algorithm to a probabilistic approach using hidden Markov models which allows us more accurate transcript detection. In addition, we develop a novel clustering algorithm for gene expression data which incorporates error information about gene expression measurements. The clustering algorithm is motivated by the repeat gene expression measurements from oligonucleotide microarrays which enable us to estimate the error of these measurements. We find that the high density of oligonucleotide microarrays and the specificity of the short oligonucleotide probes on these microarrays provide data which enables us to extract new biological understanding of an organism's transcriptome.