Prototype Based Classification in Bioinformatics

Prototype Based Classification in Bioinformatics

Frank-M. Schleif (University of Leipzig, Germany), Thomas Villmann (University of Leipzig, Germany) and Barbara Hammer (Technical University of Clausthal, Germany)
DOI: 10.4018/978-1-60960-561-2.ch217

Abstract

Bioinformatics has become an important tool to support clinical and biological research and the analysis of functional data, is a common task in bioinformatics (Schleif, 2006). Gene analysis in form of micro array analysis (Schena, 1995) and protein analysis (Twyman, 2004) are the most important fields leading to multiple sub omics-disciplines like pharmacogenomics, glycoproteomics or metabolomics. Measurements of such studies are high dimensional functional data with few samples for specific problems (Pusch, 2005). This leads to new challenges in the data analysis. Spectra of mass spectrometric measurements are such functional data requiring an appropriate analysis (Schleif, 2006). Here we focus on the determination of classification models for such data. In general, the spectra are transformed into a vector space followed by training a classifier (Haykin, 1999). Hereby the functional nature of the data is typically lost. We present a method which takes this specific data aspects into account. A wavelet encoding (Mallat, 1999) is applied onto the spectral data leading to a compact functional representation. Subsequently the Supervised Neural Gas classifier (Hammer, 2005) is applied, capable to handle functional metrics as introduced by Lee & Verleysen (Lee, 2005). This allows the classifier to utilize the functional nature of the data in the modelling process. The presented method is applied to clinical proteome data showing good results and can be used as a bioinformatics method for biomarker discovery.
Chapter Preview
Top

Prototype Based Analysis In Clinical Proteomics

Here we focus on classification models. A powerful tool to achieve such models with high generalization abilities is available with the prototype based Supervised Neural Gas algorithm (SNG) (Villmann, 2002). Like all nearest prototype classifier algorithms, SNG heavily relies on the data metric d, usually the standard Euclidean metric. For high-dimensional data as they occur in proteomic patterns, this choice is not adequate due to two reasons: first, the functional nature of the data should be kept as far as possible. Second the noise present in the data set accumulates and likely disrupts the classification when taking a standard Euclidean approach. A functional representation of the data with respect to the used metric and a weighting or pruning of especially (priory not known) irrelevant function parts of the inputs, would be desirable. We focus on a functional distance measure as recently proposed in (Lee, 2005) referred as functional metric. Additionally a feature selection is applied based on a statistical pre-analysis of the data. Hereby a discriminative data representation is necessary. The extraction of such discriminant features is crucial for spectral data and typically done by a parametric peak picking procedure (Schleif, 2006). This peak picking is often spot of criticism, because peaks may be insufficiently detected and the functional nature of the data is partially lost. To avoid these difficulties we focus on a wavelet encoding. The obtained wavelet coefficients are sufficient to reconstruct the signal, still containing all relevant information of the spectra, but are typically more complex and hence a robust data analysis approach is needed. The paper is structured as follows: first the bioinformatics methods are presented. Subsequently the clinical data are described and the introduced methods are applied in the analysis of the proteome spectra. The introduced method aims on a replacement of the classical three step procedure of denoising, peak picking and feature extraction by means of a compact wavelet encoding which gives a more natural representation of the signal.

Complete Chapter List

Search this Book:
Reset