Search

We present a set of coherent methods for the nearly automatic creation of 3D geometric models from large stacks of images of histological sections. Three-dimensional surface models facilitate the visual analysis of 3D anatomy. They also form a basis for standardized anatomical atlases that allow researchers to integrate, accumulate and associate heterogeneous experimental information, like functional or gene-expression data, with spatial or even spatio-temporal reference. Models are created by performing the following steps: image stitching, slice alignment, elastic registration, image segmentation and surface reconstruction. The proposed methods are to a large extent automatic and robust against inevitably occurring imaging artifacts. The option of interactive control at most stages of the modeling process complements automatic methods. [ABSTRACT FROM AUTHOR]

Clustering tasks occur for various different application domains including very large data streams e.g. for robotics and life science, different data formats such as graphs and profiles, and a multitude of different objectives ranging from statistical motivations to data driven quantization errors. Thus, there is a need for efficient any-time self-adaptive models and implementations. The focus of this contribution is on clustering algorithms inspired by biological paradigms which allow to transfer ideas of organic computing to the important task of efficient clustering. We discuss existing methods of adaptivity and point out a taxonomy according to which adaptivity can take place. Afterwards, we develop general perspectives for an efficient self-adaptivity of self-organizing clustering. [ABSTRACT FROM AUTHOR]

For high-throughput screening of genetically modified plant cells, a system for the automatic analysis of huge collections of microscope images is needed to decide whether the cells are infected with fungi or not. To study the potential of feature based classification for this application, we compare different classifiers (kNN, SVM, MLP, LVQ) combined with several feature reduction techniques (PCA, LDA, Mutual Information, Fisher Discriminant Ratio, Recursive Feature Elimination). We achieve a significantly higher classification accuracy using a reduced feature vector instead of the full length feature vector. [ABSTRACT FROM AUTHOR]

Multidimensional Scaling (MDS) is a powerful dimension reduction technique for embedding high-dimensional data into a low-dimensional target space. Thereby, the distance relationships in the source are reconstructed in the target space as best as possible according to a given embedding criterion. Here, a new stress function with intuitive properties and a very good convergence behavior is presented. Optimization is combined with an efficient implementation for calculating dynamic distance matrix correlations, and the implementation can be transferred to other related algorithms. The suitability of the proposed MDS for high-throughput data (HiT-MDS) is studied in applications to macroarray analysis for up to 12,000 genes. Keywords:Multi-dimensional scaling, clustering, gene expression analysis. [ABSTRACT FROM AUTHOR]

Biomedical High-Throughput Screening (HTS) requires specific properties of image compression. Particularly especially when archiving a huge number of images of one specific experiment the time factor is often rather secondary, and other features like lossless compression and a high compression ratio are much more important. Due to the similarity of all images within one experiment series, a content based compression seems to be especially applicable. Biologically inspired techniques, particularly Artificial Neural Networks (ANN) are an interesting and innovative tool for adaptive intelligent image compression, although a couple of promising non-neural alternatives, such as CALIC or JP EG2000 have become available. [ABSTRACT FROM AUTHOR]

Splice site recognition is an important subproblem of de novo gene finding, splice junctions constituting the boundary between coding and non-coding regions in eukaryotic DNA. The availability of large amounts of sequenced DNA makes the development of fast and reliable tools for automatic identi.cation of important functional regions of DNA necessary. We present a prototype based pattern recognition tool trained for automatic donor and acceptor recognition. The developed classification model is very sparse and allows fast identification of splice sites. The method is compared with a recent model based on support vector machines on two publicly available data sets, a well known benchmark from the UCIrepository for human DNA [6] and a large dataset containing DNA of C.elegans. Our method shows competitive results and the achieved model is much sparser.The program is available at http://www.informatik.uni-osnabrueck.de/lnm/upload/ [ABSTRACT FROM AUTHOR]

The use of gene chips and microarrays for measuring gene expression is becoming widespread and is producing enormous amounts of data. With increasing numbers of datasets becoming available, the need grows for well-defined, robust and interpretable methods to mine and extract knowledge from these datasets. There is currently a lot of uncertainty as to which computational and statistical methods to adopt, mainly because of the new challenges with regard to high dimensionality that gene expression data presents to the data mining community. There is a tendency for increasingly complex methods for dimensionality reduction to be proposed that are difficult to interpret. Results produced by these methods are also difficult to reproduce by other researchers. We evaluate the application of single layer, feedforward backpropagation artificial neural networks for reducing the dimensionality of both discrete and continuous gene expression data. Such networks also allow for the extraction of classification rules from the reduced data set. We demonstrate how ‘supergenes' can be extracted from combined gene expression datasets using our method. [ABSTRACT FROM AUTHOR]

Microarrays (Schena et al. 1995) are also called gene chips or DNA chips. On a microarray chip, there are thousands of spots. Each spot contains the clone of a gene from one specific tissue. At the same time, some mRNA samples are labeled with two different kinds of dyes, for example,Cy5 (red) and Cy3 (blue). After that, the mRNA samples will be put on the chip and interact with the genes on the chip. This process is called hybridization. After hybridization has finished, the color of each spot on the chip will change. The image of the chip will be scanned out. This image reflects the characteristics of the tissue at the molecular level. If we make microarrays for different tissues, biological and biomedical researchers are able to compare the difference of those tissues at the molecular level. Figure 1 is a description of the process of making microarrays. [ABSTRACT FROM AUTHOR]

Currently, cancer and other complex pathologies are analyzed mainly by morphological classification. In the past few decades there have been dramatic improvements, adding many sophisticated methods to the range of available diagnostic tools, but the traditional approaches are still in widespread usage. However, some serious limitations are known to affect these methodologies. For instance, different cancer types and clinical courses, with different response to treatments, can manifest themselves with undistinguishable appearances, not only at morphological inspection, but also from the immunophenotyping, biochemical, and cytogenetic profiles. [ABSTRACT FROM AUTHOR]

Microarray technology is having a significant impact in the biological and medical sciences and class prediction will play an increasingly important role in the use and interpretation of microarray data. For example, classifiers could be constructed indicating the detailed subtype of a disease, its expected progression and the best treatment strategy. In this chapter we outline the main stages involved in the development of a successful class predictor for microarray datasets, including data normalisation, the different classifiers which can be used, different feature selection strategies and a method for determining how much data is required for a classification task given an initial sample set. We illustrate this process with both public domain datasets and a new dataset for predicting relapse versus non-relapse for a paediatric tumour. [ABSTRACT FROM AUTHOR]

In molecular biology, functions are produced by a set of macromolecules that interact at different levels. Genes and their products, proteins, participate to regulatory networks that control the response of the cell to external input signals. One of the most important challenge to biologists is undoubtedly to understand the mechanisms that govern this regulation, and to identify among a set of genes which play a regulator role and which are regulated. While the problem used to be approached by a gene to gene approach, this is changed significantly by the development of microarray technology. Expression of thousands of genes of a given organism or a given tissue can now be measured simultaneously on the same chip. This revolution opens a large avenue for research on reconstruction of gene regulatory networks from experimental data. [ABSTRACT FROM AUTHOR]

When a yeast cell is challenged by a rapid change in the conditions, be it temperature, osmolarity, pH, nutrient or other, it starts a genome stress response program. Survival of especially single-cell organisms depends on their ability to adapt to the environmental changes and therefore stress response has received much attention. In the budding yeast Saccharomyces cerevisiae several hundred genes out of about 6500 present in the genome have previously been found involved in a stereotyped stress response pattern. Hierarchical clustering techniques applied to gene expression measurements have also previously identified a subset of genes termed common environmental stress response (CESR) or common environmental response (CER) genes, that respond in the same way in a variety of environmental conditions. There is evidence from two different sets of experiments that many of these genes are regulated by the same Msn2p and Msn4p transcription factor pair. We have extended the study by in silico data mining using a new supervised discriminative clustering (DC) technique, which directly searches for responses potentially regulated by the Msn2/4p factors. We observed a cluster of CESR/CER genes, comparable to those previously found and potentially regulated by Msn2/4p. The results of discriminative clustering both support the viability of the technique in supervised gene expression clustering and yield new insights into genomic stress response. [ABSTRACT FROM AUTHOR]

Based on the Human Genome Project, the new interdisciplinary subject of bioinformatics has become an important research topic during the last decade. A catalytic element of this process is that the methods of molecular biology (DNA sequencing, proteomics etc.) allow the automated generation of data from cellular components. Based on this technology, robots are able to sequence small genomes in a few weeks. Moreover, the semi-automatic assembly and annotation of the sequence data can only be done using the methods of computer science. To handle massive amount of data using hardware and software is one reason for the actual success of bioinformatics (Collado-Vides and Hofestädt 2002). Today, besides the genome, protein and pathway data, a new domain of data is arising - the so-called proteomic project, which allows the identi.cation of speci.c protein pro.les in concentration and location. [ABSTRACT FROM AUTHOR]