Institutes outside Greece
Statistical methods for data mining in genomics databases: gene set enrichment analysis
Abstract
dc:descriptionIn different applications, such as cancer research, it is of particular interest to identify differentially expressed genes and/or groups of genes. This need has led to the development of many statistical methods, adapted to the treatment of genomic data. This work focuses on methods comparing a vector of numeric values, such as expression levels attached to each gene in the human genome, to a given set of genes, known to be associated for instance to some type of cancer, cellular function, or biological process. Acquisition methods for numeric data vectors are discussed, and a set of R functions realizing the main formatting operations is proposed. An overview of the existing statistical tests, from single-gene analysis to gene set analysis is given. Among those, Gene Set Enrichment Analysis (GSEA) is probably the most basic tool for genomic data treatment. However, from a statistical point of view, the centering of its test statistic does not allow the derivation of asymptotic results. Moreover, the calculation of p-values is a very time-consuming, low precision procedure. A test statistic with a different centering is proposed. Under the null hypothesis, the convergence in distribution of the new test statistic is proved, using the theory of empirical processes. The limiting distribution needs to be computed only once, and can then be used for computing the p-values of many different gene sets. This results in very large savings in computing time. The test defined in this way has been called Weighted Kolmogorov Smirnov (WKS) test, since it can be viewed as a generalization of the classical Kolmogorov-Smirnov goodness-of-fit test. Next, a different problem is addressed: the large number of false discoveries from existing methods. Based on a statistical study of several databases of gene sets, coupled with another study of a large number of expression vectors, an explanation is proposed: the null hypothesis of existing methods, which is that all genes have the same probability to be included in a gene set, is far from true in practice. A generalization of the WKS test, called Doubly Weighted Kolmogorov Smirnov (DWKS) test is proposed: it is based on a differential weighting of the genes, taking into account their relative frequencies in real data. The two tests (WKS and DWKS) have been applied both to simulated and real data, and compared with other existing procedures. Using expression data from the GEO repository, tested against the MSig Databases, a comparison between the GSEA test and the new procedures has been conducted. Our conclusion is that, beyond their mathematical and algorithmic advantages, the WKS and DWKS tests could be more informative in many cases, than the classical GSEA test and efficiently address the issues that have led to their construction.
Degree
thesis:*- Grantor dc:publisher
- Institutes outside Greece
- Year dc:date
- 2015
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Charmpi, Konstantina
- Χαρμπή, Κωνσταντίνα
Subjects
dc:subject × 14Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/55672