The final public oral examination of Morten Kloster will take place on Friday, October 8, 2004 at 9:30 a.m. in Room 202. The examining committee will consist of Professors M. Aizenman, R. Austin and N. Wingreen (Molecular Biology). Any other members of the University wishing to attend the examination may do so.

The thesis of Morten Kloster, entitled Self-organized criticality, competitive evolution and analysis of gene-expression data, has been placed on deposit.

Any member of the University wishing to read the thesis may do so. Any objections should be submitted to me in writing. The principal advisor for this work was Dr. Chao Tang of NEC.

ABSTRACT

This dissertation contains work on three separate topics. (I) A directed sandpile model with stochastic toppling rules is introduced, and its asymptotic behavior is solved in all dimensions. The model clearly belongs to a dierent universality class from its counterpart with deterministic toppling rules, previously solved by Dhar and Ramaswamy. (II) The in vitro evolution of a DNA sequence via binding to a transcription factor is considered theoretically and through large-scale, realistic simulations. It is shown that the evolution behavior is qualitatively dierent in dierent parameter regimes, and in each regime we find analytical estimates that agree well with simulation results. I then introduce a more abstract, general theory of competitive evolution, and show that through the assumption on translational invariance, the mean-field theory of such evolution can be reduced to a simple, universal set of equations. The simplicity of this formulation allows simple explanations and accurate quantitative predictions, including corrections when the assumptions of translational invariance and mean field are violated. (III) The use of gene microchips has enabled a rapid accumulation of gene-expression data. One of the major challenges of analyzing this data is the diversity, in both size and signal strength, of the various modules in the gene regulatory networks of organisms. Based on the Iterative Signature Algorithm [S. Bergmann, J. Ihmels, and N. Barkai, Phys. Rev. E 67, 031902 (2002)], we present an algorithmthe Progressive Iterative Signature Algorithm (PISA)that, by sequentially eliminating modules, allows unsupervised identification of both large and small regulatory modules. PISA is shown to perform well when applied to a large set of yeast gene-expression data; in particular, using the Gene Ontology database as a reference, it is seen to much better identify regulatory modules than methods based on high-throughput transcription-factor binding experiments or on comparative genomics.

Daniel Marlow
Chair, Dept. of Physics