Computational modeling of cancer etiology and progression using neural networks and genetic cellular automata /by Armand Bankhead III.

Cancer is a pathological state characterized by abnormally increased cell reproduction and survival. Cancer is caused by mutations to tumor suppressor and apoptosis genes which inhibit cellular reproduction and proto-oncogenes that activate reproduction. The specific genes that are mutated vary between individuals and the originating tissue. It is difficult to diagnose when and where in the body these mutations will occur.;Computational models provide a framework for abstraction of biochemically complex, biological systems relevant to cancer etiology and progression. Computer simulations can be implemented to run much faster and in a more cost-effective manner than biological experiments while producing a greater amount of data. In silico models allow us to ask questions about features that would not be possible with classical in vitro experiments, such as lethal genotypes or unidentified genes. Four research papers are included in this dissertation that present novel and useful computational models of cancerous behavior.;Genes interact with other genes and gene products in the form of genetic networks. We simulate the well established G2/M genetic network by implementing it as a neural network. This neural network is trained to reproduce in vivo mouse knockout data by disabling nodes in the neural network. The trained neural network is analyzed to quantify the importance of knockout genes p53, BRCA1, and ATM, to mammary cancer susceptibility.;We use cellular automata (CA) to model a specific form of breast cancer called ductal carcinoma in situ (DCIS). We present a novel extension to CA design by implementing CA rules as heritable genes that are subject to mutation. We also implement a newly discovered progenitor hierarchy that allows only progenitor cell types to reproduce. To examine the effect of progenitor hierarchy structure on cancer incidence, genetic heterogeneity, and cancer growth, we use several hierarchical structures. We also examine the effects of hereditary genetic predisposition by running simulations with and without initial mutations.;The research presented uses established computational paradigms established for decades. These models have been extended in novel and biologically realistic ways to answer fundamental questions regarding cancer etiology and progression.Thesis (Ph. D., Bioinformatics and Computational Biology)--University of Idaho, December 2006