Data mining, neural nets, trees — Problems 2 and 3 of Genetic Analysis Workshop 15

Abstract

Genome-wide association studies using thousands to hundreds of thousands of single nucleotide polymorphism (SNP) markers and region-wide association studies using a dense panel of SNPs are already in use to identify disease susceptibility genes and to predict disease risk in individuals. Because these tasks become increasingly important, three different data sets were provided for the Genetic Analysis Workshop 15, thus allowing examination of various novel and existing data mining methods for both classification and identification of disease susceptibility genes, gene by gene or gene by environment interaction. The approach most often applied in this presentation group was random forests because of its simplicity, elegance, and robustness. It was used for prediction and for screening for interesting SNPs in a first step. The logistic tree with unbiased selection approach appeared to be an interesting alternative to efficiently select interesting SNPs. Machine learning, specifically ensemble methods, might be useful as pre-screening tools for large-scale association studies because they can be less prone to overfitting, can be less computer processor time intensive, can easily include pair-wise and higher-order interactions compared with standard statistical approaches and can also have a high capability for classification. However, improved implementations that are able to deal with hundreds of thousands of SNPs at a time are required.

Summary of contributions to GAW15 Group 13: candidate gene association studies

“…Brinza et al [personal communication] applied a complimentary greedy search algorithm, a machine learning approach developed by his group [Brinza and Zelikovsky, 2006], to identify SNPs associated with RA, and also to predict susceptibility of the tested genotype to RA subjects using the Problem 2 data set. Their paper is discussed in more detail in the Group 6 summary paper [Ziegler et al, 2007]. Qin et al [2007] developed a graphical display tool called SIMLA-PLOT for visualizing different ways in which continuous covariates may influence the genotypespecific risk for complex human diseases.…”

Patient-centered yes/no prognosis using learning machines

“…Applications of random forests in medical research have mostly focused on the classification of genetic data (e.g., Schwarz et al, 2007;Schwender et al, 2004; for an overview, see Ziegler et al (2007)). As the name implies, the basic units of this method are trees, and it utilises a combination of manipulating the training cases together with introducing an additional element of randomness.…”

“…The second importance measure is a generalisation of the Gini index from a single tree to a forest. The basic idea of this importance measure is to contrast the impurity of a tree with and without the feature of interest being included in the tree; for details, see Ziegler et al (2007). If the estimated importance of all features can be assumed to be independent from tree to tree, a standard error of the importance can be computed in a usual way so that asymptotic confidence intervals assuming normality can be calculated (Lin et al, 2004).…”

“…For example, one might use random forest on a full set of features to identify the most important variables and then send this small list to a logistic regression model which then lends itself readily to interpretability (Schwarz et al, 2007). In detail, the following strategy seems to be reasonable (see, e.g., Ziegler et al, 2007):…”

Data mining of RNA expression and DNA genotype data: Presentation Group 5 contributions to Genetic Analysis Workshop 15

“…Although the approaches taken and goals proposed are very different, there are common themes in the approaches as well as a remarkable level of confirmation of some results. Although such important techniques as random forest [Breiman, 2001], boosting [Schapire, 1990] and ensemble approaches [Dietterich, 2000] used extensively in the data mining analyses for the other problems [Ziegler et al, 2007], these techniques were not applied in this group. Table I summarizes the 13 papers, indicating common themes among many of them.…”