Catálogo de publicaciones - libros

Important issues in bioinformatics are the difficulties for non computer experts to use bioinformatics tools, the transparent access to large biological data sets, and the exploitation of large computing power. Moreover, often such tools and databases are developed by independent groups, so the task of enabling their composition and cooperation is even more difficult. Integrating Computational Grid and Web Services technologies can be a key solution to simplify interaction between bioinformatics tools and biological databases. This paper presents ProGenGrid (Proteomics & Genomics Grid), a distributed and ubiquitous grid environment, accessible through the web, for supporting “” experiments in bioinformatics.

Head and neck squamous cell carcinoma (HNSCC) has already been proved to be linked with smoking and alcohol drinking habits. However the individual risk could be modified by genetic polymorphisms of enzymes involved in the metabolism of tobacco carcinogens and in the DNA repair mechanisms. To study this relationship, a data set comprising clinical (age, smoke, alcohol) and genetic data (the genetic polymorphism of 11 genes) was built; an XCS system was then developed in order to analyze it. XCS appears well suited to this problem since it can seamlessly accept missing values, and be adapted to deal with different data types (real, integer, and class). Moreover, it produces human-readable rules - which is fundamental in order to make the system useful to physicians. First results showed interesting rules, suggesting that this approach is viable and deserves deeper research.

Mass Spectrometry (MS) can be used as a detector in High Performance Liquid Chromatography (HPLC) systems or as a tool for direct protein/peptides profiling from biological samples. Data Mining (DM) is the semi-automated extraction of patterns representing knowledge implicitly stored in large databases. The combined use of MS with DM is a novel approach in proteomic pattern analysis and is emerging as an effective method for the early diagnosis of diseases. We describe the workflow of a proteomic experiment for early detection of cancer which combines MS and DM, giving details of sample treatment and preparation, MS data generation, MS data preprocessing, data clustering and classification.

The bio-molecular diagnosis of malignancies represents a difficult learning task, because of the high dimensionality and low cardinality of the data. Many supervised learning techniques, among them support vector machines, have been experimented, using also feature selection methods to reduce the dimensionality of the data. In alternative to feature selection methods, we proposed to apply random subspace ensembles, reducing the dimensionality of the data by randomly sampling subsets of features and improving accuracy by aggregating the resulting base classifiers. In this paper we experiment the combination of random subspace with feature selection methods, showing preliminary experimental results that seem to confirm the effectiveness of the proposed approach.

In this paper we describe and compare two different methods to reduce the cardinality of the set of candidates nodules, characterized by an high sensitivity ratio, and extracted from PA chest radiographs by a fully automatized method. The methods are a rule based system and a feed-forward neural network trained by back-propagation. Both the systems allow to recognize almost the 75% of false positives without losing any true positives.

We develop and test a new hierarchical approach for the prediction of protein structure. An algorithm is described to assemble the 3D fold of a protein starting from its secondary structure and -sheet topology. Reconstruction is carried out by energy minimization of a reduced protein model, where -partners are derived from appropriate distance constraints imposed by the knowledge of -sheet motifs. Additional constraints are imposed in the () torsion space from secondary structure knowledge. Experiments show how the proposed procedure proves to be a reliable and fast predictive approach for a large fraction of proteins of interest. Arrangements of -sheets are predicted with special recursive neural networks architectures. We first present a unifying framework for description of a large class of contextual recursive models and then show how it is possible to solve the problem at some extent of success.

The Human Leukocyte Antigen (HLA) region is a part of genome which spans over 4 Mbases of DNA. The HLA system is strongly connected to immunological response and its compatibility between tissues is critical in transplantation. We have developed an application of oligonucleotide microarrays to HLA typing. In this paper we present a method based on a fuzzy system which interactively supports the user in analyzing the hybridization results, speeding-up the decision process moving from raw array data obtained from the scanner to their interpretation (genotyping). The two-level procedure starts with evaluation of spot activity, then it estimates probe hybridization levels from activity levels. The method is designed for being readily usable by the biologist, by adopting fuzzy linguistic variables which are familiar to the user and by featuring a standard and complete graphical interface.

In this paper we describe the use of a correlation clustering algorithm [Chaitanya, 2004] to group expression level of genes in a microarray dataset. The clustering problem is formalized as a semi-defined optimization program, based on the correlation provided by two quantities, respectively related to an agreement and a disagreement between a pair of genes. We also intend to validate the role of the correlation clustering algorithm by comparing the results with a support vectors clustering approach [Ben-Hur et al., 2001] that is demonstrated to perform well for many applications.

Several gene selection methods have been proposed to identify sets of genes related to a particular disease or to a particular functional status of the tissue. An open problem with gene selection methods consists in evaluating their performance; since we usually know only a smell subset of the genes involved in the onset of a status, and many times no relevant genes are known “a priori”. We propose an artificial system, based on modeling gene expression signatures, to generate synthetic gene expression data for validating gene selection methods. Comparison between gene selection methods using data generated through the artificial model are performed and preliminary results are reported.

Gene-expression microarrays make it possible to simultaneously measure the rate at which a cell or tissue is expressing each of its thousands of genes. One can use these comprehensive snapshots of biological activity to infer regulatory pathways in cells, identify novel targets for drug design, and improve diagnosis, prognosis, and treatment planning for those suffering from disease. However, the amount of data this new technology produces is more than one can manually analyze. Hence, the need for automated analysis of microarray data offers an opportunity for machine learning to have a significant impact on biology and medicine. Probabilistic Principal Surfaces defines a unified theoretical framework for nonlinear latent variable models embracing the Generative Topographic Mapping as a special case. This article describes the use of PPS for the analysis of yeast gene expression levels from microarray chips showing its effectiveness for high-D data visualization and clustering.