| Summary: | 碩士 === 國立陽明大學 === 生物醫學資訊研究所 === 106 === That cancer progression is related to driver mutations has been recorded, and single gene mutations do not contribute to tumorigenesis. The combination of mutated genes that drive normal cells to gradually transform into cancer cells is also called the driver gene sets. Some literatures (e.g., Dendrix and MDPFinder) apply subnetwork-based approach to identify driver gene sets through different algorithms, and the searching metrics must satisfy two properties: (1) High exclusivity: which means that nearly all patients have no more than one mutation in the driver gene set. (2) High coverage: which means that most patients have at least one mutation in the driver gene set.
Our research designed a quadratic objective function based on gradient descent approach, which we take as the concept of optimization, in order to identify driver gene sets. We first adjust two parameters (C_0 and C_1^' s range) to test different size of datasets, and used permutation test to gain statistical significance of each gene set, and then compare Dendrix with MDPFinder in terms of diversity of results, besides obtaining similar results in right area under pareto curve, we can get our unique results in left area under pareto curve, that is to say, different combination of genes can be searched by our method. We also discovered that different long tail distribution will affect the diversity of results. In terms of biological annotation, we can map driver gene sets into cancer-related pathways, and also identify survival-related driver gene sets by survival analysis. Our main goal is to apply driver gene sets to the development of precision medicine and look for the corresponding cancer treatment drugs.
|
|---|
