Gene expression data analysis using novel methods: Predicting time delayed correlations and evolutionarily conserved functional modules

Declaration I hereby declare that the dissertation entitled " Gene expression data analysis using novel methods: Predicting time delayed correlations and evolutionarily conserved functional modules " submitted to the Department of Biology, Philipps-Universität, Marburg is the original and independent work carried out by me under the guidance of the PhD committee, and the dissertation is not formed previously on the basis of any award of Degree, Diploma or other similar titles. (Date and Place) (Rajarajeswari Balasubramaniyan) On action alone be thy interest, Never on its fruits. Let not the fruits of action be thy motive, Nor be thy attachment to inaction. Synopsis I Synopsis Microarray technology enables the study of gene expression on a large scale. One of the main challenges has been to devise methods to cluster genes that share similar expression profiles. In gene expression time courses, a particular gene may encode transcription factor and thus controlling several genes downstream; in this case, the gene expression profiles may be staggered, indicating a time-delayed response in transcription of the later genes. The standard clustering algorithms consider gene expression profiles in a global way, thus often ignoring such local time-delayed correlations. We have developed novel methods to capture time-delayed correlations between expression profiles: (1) A method using dynamic programming and (2) CLARITY, an algorithm that uses a local shape based similarity measure to predict time-delayed correlations and local correlations. We used CLARITY on a dataset describing the change in gene expression during the mitotic cell cycle in Saccharomyces cerevisiae. The obtained clusters were significantly enriched with genes that share similar functions, reflecting the fact that genes with a similar function are often co-regulated and thus co-expressed. Time-shifted as well as local correlations could also be predicted using CLARITY. In datasets, where the expression profiles of independent experiments are compared, the standard clustering algorithms often cluster according to all conditions, considering all genes. This increases the background noise and can lead to the missing of genes that change the expression only under particular conditions. We have employed a genetic algorithm based module predictor that is capable to identify group of genes that change their expression only in a subset of conditions. With the aim of supplementing the Ustilago maydis genome annotation, we have used the module prediction algorithm on various independent datasets from Ustilago maydis. The predicted modules were cross-referenced in various Saccharomyces cerevisiae datasets to check its evolutionarily conservation between …