Estimation of Relevant Variables on High-Dimensional Biological Patterns Using Iterated Weighted Kernel Functions

Abstract

BackgroundThe analysis of complex proteomic and genomic profiles involves the identification of significant markers within a set of hundreds or even thousands of variables that represent a high-dimensional problem space. The occurrence of noise, redundancy or combinatorial interactions in the profile makes the selection of relevant variables harder.Methodology/Principal FindingsHere we propose a method to select variables based on estimated relevance to hidden patterns. Our method combines a weighted-kernel discriminant with an iterative stochastic probability estimation algorithm to discover the relevance distribution over the set of variables. We verified the ability of our method to select predefined relevant variables in synthetic proteome-like data and then assessed its performance on biological high-dimensional problems. Experiments were run on serum proteomic datasets of infectious diseases. The resulting variable subsets achieved classification accuracies of 99% on Human African Trypanosomiasis, 91% on Tuberculosis, and 91% on Malaria serum proteomic profiles with fewer than 20% of variables selected. Our method scaled-up to dimensionalities of much higher orders of magnitude as shown with gene expression microarray datasets in which we obtained classification accuracies close to 90% with fewer than 1% of the total number of variables.ConclusionsOur method consistently found relevant variables attaining high classification accuracies across synthetic and biological datasets. Notably, it yielded very compact subsets compared to the original number of variables, which should simplify downstream biological experimentation.

Data-Mining Techniques to Classify Microarray Gene Expression Data Using Gene Selection by SVD and Information Gain

“…We approve that our results are similar with other studies. For instance, Rojas-Galeano et al [27] gets 88% with their kernel-based algorithm. Futschik et al [2] reported that Colon dataset gives lower accuracy compared to other datasets.…”

Goldenberry

“…The aim there is to select an optimal subset of relevant variables for a classification problem; the relevant found subset can be further analysed by experts for specific purposes (in biology for example, features may represent over-expressed gene activity due to an illness condition). In the so-called wrapper scheme of the problem [18] the classifier is enabled to incorporate the selection mechanism during the learning stage: a weighted kernel classifier, for example, may use an UMDA component to define relevance coefficients for the variables and then use them as input for the kernel machine (that is the approach taken in [16]). Currently these type of mechanisms are hidden in the current implementation of classifier or clustering components of Orange.…”

Discovering feature relevancy and dependency by kernel-guided probabilistic model-building evolution

“…variables) in order to modulate their contribution to the total computation [14, 26, 27]. The weighted RBF and weighted polynomial kernels are then defined as , and , respectively.…”

“…Previous studies considered combining classical and probabilistic genetic algorithms with weighted kernel classifiers for relevancy–only discovery [14, 15]; our contribution in this paper is to extend those approaches to take advantage of the discrimination power of a weighted kernel classifier to guide the search for a probabilistic model that simultaneously estimates marginal and interacting effects among the features in a discrimination problem.…”