PhD Thesis Defence – Sajid Marhon

PhD candidate Sajid Marhon will defend his thesis "A New, Model-Independent, Spectrum-Based Gene Prediction Technique" on April 10, 2015, at 9:00pm in Reynolds 219.

Title

A New, Model-Independent, Spectrum-Based Gene Prediction Technique

Abstract

Detecting protein-coding regions is a fundamental step in genome analysis. This step is a precursor to analyzing protein sequences. Different techniques have been proposed for detecting protein-coding regions in DNA sequences. Popular techniques use probabilistic models such as Hidden Markov Models (HMMs) that depend on homology information (a learning model) in the analysis of DNA sequences, and this makes the application of these techniques restricted to species that have homologs. This approach is classified as a model-dependent method. Digital Signal Processing (DSP)-based techniques, which rely on the spectrum analysis of DNA sequences, extract the period-3 spectrum of DNA sequences to detect protein-coding regions. DSP-based techniques are classified as model-independent methods. In this research, I propose a new, DSP-based technique that overcomes the limitation in the prediction accuracy of the current DSP-based techniques and the problem of application specificity of learning-based techniques. In my technique, I propose four improvements in different stages of the gene prediction process: (1) a dynamic representation scheme, (2) an efficient method for computing nucleotide distribution variance, (3) post-processing to attenuate background noise and detect spectrum peaks without requiring an experimental threshold, and (4) a new wide-range wavelet window. My experimental results show that my technique outperforms all other popular DSP-based techniques. In addition, the comparison of the results of the proposed technique with the popular HMMgene technique shows that my technique performs better on the novel gene detection problem. I believe that this is an area of research that has been underemphasized and deserves additional attention.

Advisors: Stefan Kremer