Band 7, Heft 3 - Special Issue: 6th International Symposium on Integrative Bioinformatics, Cambridge, United Kingdom, 2010 / Guest Editors: Matthias Lange, Stephan Weise, Uwe Scholz, Paul Verrier, Chris Rawlings

Visualization is pivotal for gaining insight in systems biology data. As the size and complexity of datasets and supplemental information increases, an efficient, integrated framework for general and specialized views is necessary. MAYDAY is an application for analysis and visualization of general ‘omics’ data. It follows a trifold approach for data visualization, consisting of flexible data preprocessing, highly customizable data perspective plots for general purpose visualization and systems based plots. Here, we introduce two new systems biology visualization tools for MAYDAY. Efficiently implemented genomic viewers allow the display of variables associated with genomic locations. Multiple variables can be viewed using our new track-based ChromeTracks tool. A functional perspective is provided by visualizing metabolic pathways either in KEGG or BioPax format. Multiple options of displaying pathway components are available, including Systems Biology Graphical Notation (SBGN) glyphs. Furthermore, pathways can be viewed together with gene expression data either as heatmaps or profiles. We apply our tools to two ‘omics’ datasets of Pseudomonas aeruginosa. The general analysis and visualization tools of MAYDAY as well as our ChromeTracks viewer are applied to a transcriptome dataset. We furthermore integrate this dataset with a metabolome dataset and compare the activity of amino acid degradation pathways between these two datasets, by visually enhancing the pathway diagrams produced by MAYDAY.

Drug development is expensive and prone to failure. It is potentially much less risky and expensive to reuse a drug developed for one condition for treating a second disease, than it is to develop an entirely new compound. Systematic approaches to drug repositioning are needed to increase throughput and find candidates more reliably. Here we address this need with an integrated systems biology dataset, developed using the Ondex data integration platform, for the in silico discovery of new drug repositioning candidates. We demonstrate that the information in this dataset allows known repositioning examples to be discovered. We also propose a means of automating the search for new treatment indications of existing compounds.

We investigated the problem of imprecisely determined prokaryotic transcription factor (TF) binding sites (TFBSs). We found that the identification and reinvestigation of questionable binding motifs may result in improved models of these motifs. Subsequent modelbased predictions of gene regulatory interactions may be performed with increased accuracy when the TFBSs annotation underlying these models has been re-adjusted. We present MoRAine 2.0, a significantly improved version of MoRAine. It can automatically identify cases of unfavorable TFBS strand annotations and imprecisely determined TFBS positions. With release 2.0, we close the gap between reasonable running time and high accuracy. Furthermore, it requires only minimal input from the user: (1) the input TFBS sequences and (2) the length of the flanking sequences. Conclusions: MoRAine 2.0 is an easy-to-use, integrated, and publicly available web tool for the re-annotation of questionable TFBSs. It can be used online or downloaded as a stand-alone version from http://moraine.cebitec.uni-bielefeld.de.

Efficient and effective information retrieval in life sciences is one of the most pressing challenge in bioinformatics. The incredible growth of life science databases to a vast network of interconnected information systems is to the same extent a big challenge and a great chance for life science research. The knowledge found in the Web, in particular in life-science databases, are a valuable major resource. In order to bring it to the scientist desktop, it is essential to have well performing search engines. Thereby, not the response time nor the number of results is important. The most crucial factor for millions of query results is the relevance ranking. In this paper, we present a feature model for relevance ranking in life science databases and its implementation in the LAILAPS search engine. Motivated by the observation of user behavior during their inspection of search engine result, we condensed a set of 9 relevance discriminating features. These features are intuitively used by scientists, who briefly screen database entries for potential relevance. The features are both sufficient to estimate the potential relevance, and efficiently quantifiable. The derivation of a relevance prediction function that computes the relevance from this features constitutes a regression problem. To solve this problem, we used artificial neural networks that have been trained with a reference set of relevant database entries for 19 protein queries. Supporting a flexible text index and a simple data import format, this concepts are implemented in the LAILAPS search engine. It can easily be used both as search engine for comprehensive integrated life science databases and for small in-house project databases. LAILAPS is publicly available for SWISSPROT data at http://lailaps.ipk-gatersleben.de

Numerous biomolecular data are available, but they are scattered in many databases and only some of them are curated by experts. Most available data are computationally derived and include errors and inconsistencies. Effective use of available data in order to derive new knowledge hence requires data integration and quality improvement. Many approaches for data integration have been proposed. Data warehousing seams to be the most adequate when comprehensive analysis of integrated data is required. This makes it the most suitable also to implement comprehensive quality controls on integrated data. We previously developed GFINDer (http://www.bioinformatics.polimi.it/GFINDer/), a web system that supports scientists in effectively using available information. It allows comprehensive statistical analysis and mining of functional and phenotypic annotations of gene lists, such as those identified by high-throughput biomolecular experiments. GFINDer backend is composed of a multi-organism genomic and proteomic data warehouse (GPDW). Within the GPDW, several controlled terminologies and ontologies, which describe gene and gene product related biomolecular processes, functions and phenotypes, are imported and integrated, together with their associations with genes and proteins of several organisms. In order to ease maintaining updated the GPDW and to ensure the best possible quality of data integrated in subsequent updating of the data warehouse, we developed several automatic procedures. Within them, we implemented numerous data quality control techniques to test the integrated data for a variety of possible errors and inconsistencies. Among other features, the implemented controls check data structure and completeness, ontological data consistency, ID format and evolution, unexpected data quantification values, and consistency of data from single and multiple sources. We use the implemented controls to analyze the quality of data available from several different biological databases and integrated in the GFINDer data warehouse. By doing so, we identified in these data a variety of different types of errors and inconsistencies; this enables us to ensure good quality of the data in the GFINDer data warehouse. We reported all identified data errors and inconsistencies to the curators of the original databases from where the data were retrieved, who mainly corrected them in subsequent updating of the original database. This contributed to improve the quality of the data available, in the original databases, to the whole scientific community.

Networks are used to model real-world phenomena in various domains, including systems biology. Since proteins carry out biological processes by interacting with other proteins, it is expected that cellular functions are reflected in the structure of protein-protein interaction (PPI) networks. Similarly, the topology of residue interaction graphs (RIGs) that model proteins’ 3-dimensional structure might provide insights into protein folding, stability, and function. An important step towards understanding these networks is finding an adequate network model, since models can be exploited algorithmically as well as used for predicting missing data. Evaluating the fit of a model network to the data is a formidable challenge, since network comparisons are computationally infeasible and thus have to rely on heuristics, or “network properties.” We show that it is difficult to assess the reliability of the fit of a model using any network property alone. Thus, we present an integrative approach that feeds a variety of network properties into five machine learning classifiers to predict the best-fitting network model for PPI networks and RIGs. We confirm that geometric random graphs (GEO) are the best-fitting model for RIGs. Since GEO networks model spatial relationships between objects and are thus expected to replicate well the underlying structure of spatially packed residues in a protein, the good fit of GEO to RIGs validates our approach. Additionally, we apply our approach to PPI networks and confirm that the structure of merged data sets containing both binary and co-complex data that are of high coverage and confidence is also consistent with the structure of GEO, while the structure of less complete and lower confidence data is not. Since PPI data are noisy, we test the robustness of the five classifiers to noise and show that their robustness levels differ. We demonstrate that none of the classifiers predicts noisy scale-free (SF) networks as GEO, whereas noisy GEOs can be classified as SF. Thus, it is unlikely that our approach would predict a real-world network as GEO if it had a noisy SF structure. However, it could classify the data as SF if it had a noisy GEO structure. Therefore, the structure of the PPI networks is the most consistent with the structure of a noisy GEO.

High throughput genomic studies can identify large numbers of potential candidate genes, which must be interpreted and filtered by investigators to select the best ones for further analysis. Prioritization is generally based on evidence that supports the role of a gene product in the biological process being investigated. The two most important bodies of information providing such evidence are bioinformatics databases and the scientific literature. In this paper we present an extension to the Ondex data integration framework that uses text mining techniques over Medline abstracts as a method for accessing both these bodies of evidence in a consistent way. In an example use case, we apply our method to create a knowledge base of Arabidopsis proteins implicated in plant stress response and use various scoring metrics to identify key protein-stress associations. In conclusion, we show that the additional text mining features are able to highlight proteins using the scientific literature that would not have been seen using data integration alone. Ondex is an open-source software project and can be downloaded, together with the text mining features described here, from www.ondex.org.

There is currently no standardized approach for parsing output that the numerous bioinformatical tools generate. Because the framework approach of the Bio-toolkits has some shortcomings, we searched for alternative approaches. To this end, we evaluated scanner generators for various programming languages with respect to their potential of creating standalone, small, and fast applications that can easily be delivered on any modern and many ancient operating systems. We developed sample applications that generate standard SQL database code and thereby greatly simplify the parsing work of data integration and data analysis. At the sourceforge project page the source code and some binaries for a selection of our applications are freely available at http://bioscanners.sourceforge.net.

In recent years the amount of biological data has exploded to the point where much useful information can only be extracted by complex computational analyses. Such analyses are greatly facilitated by metadata standards, both in terms of the ability to compare data originating from different sources, and in terms of exchanging data in standard forms, e.g. when running processes on a distributed computing infrastructure. However, standards thrive on stability whereas science tends to constantly move, with new methods being developed and old ones modified. Therefore maintaining both metadata standards, and all the code that is required to make them useful, is a non-trivial problem. Memops is a framework that uses an abstract definition of the metadata (described in UML) to generate internal data structures and subroutine libraries for data access (application programming interfaces - APIs - currently in Python, C and Java) and data storage (in XML files or databases). For the individual project these libraries obviate the need for writing code for input parsing, validity checking or output. Memops also ensures that the code is always internally consistent, massively reducing the need for code reorganisation. Across a scientific domain a Memops-supported data model makes it easier to support complex standards that can capture all the data produced in a scientific area, share them among all programs in a complex software pipeline, and carry them forward to deposition in an archive. The principles behind the Memops generation code will be presented, along with example applications in Nuclear Magnetic Resonance (NMR) spectroscopy and structural biology.

In this paper we evaluate the performance and usefulness of three semantic background knowledge sources for predicting synonymous anatomical terms across species boundaries. The reference sources under evaluation are UMLS, FMA-OBO and WordNet, which are applied to the anatomical ontologies of mouse and zebrafish. Our results show that the use of specialized knowledge sources leads to highly accurate predictions, verified through complete manual curation, which can be further improved by combining multiple of said sources. We argue that these three references complement each other in terms of granularity and specificity. From our results we conclude that these references can be used to create reliable ontology mappings with minimal human supervision.

Crop plants play a major role in human and animal nutrition and increasingly contribute to chemical or pharmaceutical industry and renewable resources. In order to achieve important goals, such as the improvement of growth or yield, it is indispensable to understand biological processes on a detailed level. Therefore, the well-structured management of fine-grained information about metabolic pathways is of high interest. Thus, we developed the MetaCrop information system, a manually curated repository of high quality information concerning the metabolism of crop plants. However, the data access to and flexible export of information of MetaCrop in standard exchange formats had to be improved. To automate and accelerate the data access we designed a set of web services to be integrated into external software. These web services have already been used by an add-on for the visualisation toolkit VANTED. Furthermore, we developed an export feature for the MetaCrop web interface, thus enabling the user to compose individual metabolic models using SBML.

The paper presents an ontology for the description of Drug Discovery Investigation (DDI). This has been developed through the use of a Robot Scientist “Eve”, and in consultation with industry. DDI aims to define the principle entities and the relations in the research and development phase of the drug discovery pipeline. DDI is highly transferable and extendable due to its adherence to accepted standards, and compliance with existing ontology resources. This enables DDI to be integrated with such related ontologies as the Vaccine Ontology, the Advancing Clinico-Genomic Trials on Cancer Master Ontology, etc. DDI is available at http://purl.org/ddi/wikipedia or http://purl.org/ddi/home

microRNAs are short RNA fragments that have the capacity of regulating hundreds of target gene expression. Currently, due to lack of high-throughput experimental methods for miRNA target identification, a collection of computational target prediction approaches have been developed. However, these approaches deal with different features or factors are weighted differently resulting in diverse range of predictions. The prediction accuracy remains uncertain. In this paper, three commonly used target prediction algorithms are evaluated and further integrated using algorithm combination, ranking aggregation and Bayesian Network classification. Our results revealed that each individual prediction algorithm displays its advantages as was shown on different test data sets. Among different integration strategies, the application of Bayesian Network classifier on the features calculated from multiple prediction methods significantly improved target prediction accuracy.

Mining patterns of gene expression provides a crucial approach in discovering knowledge such as finding genetic networks that underpin the embryonic development. Analysis of mining results and evaluation of their relevance in the domain remains a major concern. In this paper we describe our explorative studies in support of solutions to facilitate the analysis and interpretation of mining results. In our particular case we describe a solution that is found in the extension of the Gene Expression Management System (GEMS), i.e. an integrative framework for spatio-temporal organization of gene expression patterns of zebrafish to a framework supporting data mining, data analysis and patterns interpretation As a proof of principle, the GEMS has been equipped with data mining functionality suitable for spatio-temporal tracking, thereby generating added value to the submission of data for data mining and analysis. The analysis of the genetic networks is based on the availability of domain ontologies which dynamically provides meaning to the discovered patterns of gene expression data. Combination of data mining with the already presently available capabilities of GEMS will significantly augment current data processing and functional analysis strategies

We study, in this paper, a model for the core of the system of the Glycerophospholipid metabolism in the murine cells. It comprises the simple and enzymatic reactions of PhosphatidylEthanolamine and the PhosphatidylCholine . The model’s general structure is taken from a number of books and articles. We translate this model into a set of ordinary differential equations (ODEs), to propose a quantitative explanation of the experimental experiences and the observed results. In order to make it usable as a basis for simulations and mathematical analysis we need to make precise the various constants present in the equations but which are usually not directly accessible in the literature. In a first step we considered experimental data of rat’s liver cells obtained by NMR spectroscopy: given the values of metabolite concentrations we find appropriate parameter values which allow us to describe the system with ODEs. We have then performed several analyses using the developed model such as stability analysis. A first interesting result is the global stability of the system which was observed by simulation and then proved by mathematical arguments. A second important result is that we observe on the diagrams that the steady state for normal cells is precisely a singular point of order two, whereas tumoral cells present different characteristics; this fact has been proved for PhosphatidylEthanolamine N-Methyl transferase (PEMT), an enzyme which seems to be identified for the first time as a crucial element in the tumoral process. In a second step we applied our model to experimental data of proton HRMAS NMR spectroscopy for solid B16 melanoma and Lewis lung (3LL) 3LL carcinoma cells treated by Chloroethyl Nitrosourea (CENU). We performed a complete comparative analysis of parameters in order to learn the predictive statements to explain increases and decreases which one can observe in concentrations.

A significant part of cellular proteins undergo reversible thiol-dependent redox transitions which often control or switch protein functions. Thioredoxins and glutaredoxins constitute two key players in this redox regulatory protein network. Both interact with various categories of proteins containing reversibly oxidized cysteinyl residues. The identification of thioredoxin/glutaredoxin target proteins is a critical step in constructing the redox regulatory network of cells or subcellular compartments. Due to the scarcity of thioredoxin/glutaredoxin target protein records in the public database, a tool called Reversibly Oxidized Cysteine Detector (ROCD) is implemented here to identify potential thioredoxin/glutaredoxin target proteins computationally, so that the in silico construction of redox regulatory network may become feasible. ROCD was tested on 46 thioredoxin target proteins in plant mitochondrion, and the recall rate was 66.7% when 50% sequence identity was chosen for structural model selection. ROCD will be used to predict the thioredoxin/glutaredoxin target proteins in human liver mitochondrion for our redox regulatory network construction project. The ROCD will be developed further to provide prediction with more reliability and incorporated into biological network visualization tools as a node prediction component. This work will advance the capability of traditional database- or text mining-based method in the network construction.

The definition of a phase shift of triplet periodicity (TP) is introduced. The mathematical algorithm for detection of TP phase shift of nucleotide sequences has been developed. Gene sequences from Kegg-46 data bank were analyzed with a purpose of searching genes with a phase shift of TP. The presence of a phase shift of triplet periodicity has been shown for 318329 genes (~10% from the number of genes in Kegg-46). We suppose that shifts of the TP phase may indicate the shifts of reading frame (RF) in genes. A relationship between the phase shifts of TP and the frame shifts in genes is discussed.

Many methods for the analysis of gene expression-, protein- or metabolite-data focus on the investigation of binary relationships, while the underlying biological processes creating this data may generate relations of higher than bivariate complexity. We give a novel method ExPlanes that helps to explore certain types of ternary relationships in a statistically robust, Bayesian framework. To arrive at an characterization of the data structure contained in triplet data we investigate 2-dimensional planes being the only linear structures that cannot be inferred from projections of the data. The key part of our methodology is the definition of a robust, Bayesian plane posterior under the assumption of an invariant prior and a Gaussian error model. A numerical representation of the plane posterior can be explored interactively. Beyond this purely Bayesian approach we can use the plane posterior to construct a family of posterior-based test statistics that allow testing the data for different plane related hypotheses. To demonstrate practicability we queried triplets of metabolic data from a plant crossing experiment for the presence of plane-, line- and point-structures by using posterior-based test statistics and were able to show their distinctiveness.

Computational models in systems biology are usually characterized by a lack of reliable parameter values. This is especially true for kinetic metabolic models. Experimental data can be used to estimate these missing parameters. Different optimization techniques have been explored to solve this challenging task but none has proved to be superior to the other. In this paper we review the problem of parameter estimation in kinetic models. We focus on the suitability of four commonly used optimization techniques of parameter estimation in biochemical pathways and make a comparison between those methods. The suitability of each technique is evaluated based on the ability of converging to a solution within a reasonable amount of time. As most local optimization methods fail to arrive at a satisfactory solution we only considered the global optimization techniques. A case study of the upper part of Glycolysis consisting 15 parameters is taken as the benchmark model for evaluating these methods.

Clustering is an important approach in the analysis of biological data, and often a first step to identify interesting patterns of coexpression in gene expression data. Because of the high complexity and diversity of gene expression data, many genes cannot be easily assigned to a cluster, but even if the dissimilarity of these genes with all other gene groups is large, they will finally be forced to become member of a cluster. In this paper we show how to detect such elements, called unstable elements. We have developed an approach for iterative clustering algorithms in which unstable elements are deleted, making the iterative algorithm less dependent on initial centers. Although the approach is unsupervised, it is less likely that the clusters into which the reduced data set is subdivided contain false positives. This clustering yields a more differentiated approach for biological data, since the cluster analysis is divided into two parts: the pruned data set is divided into highly consistent clusters in an unsupervised way and the removed, unstable elements for which no meaningful cluster exists in unsupervised terms can be given a cluster with the use of biological knowledge and information about the likelihood of cluster membership. We illustrate our framework on both an artificial and real biological data set.

Traditional approaches for homology detection rely on finding sufficient similarities between protein sequences. Motivated by studies demonstrating that from non-sequence based sources of biological information, such as the secondary or tertiary molecular structure, we can extract certain types of biological knowledge when sequence-based approaches fail, we hypothesize that protein-protein interaction (PPI) network topology and protein sequence might give insights into different slices of biological information. Since proteins aggregate to perform a function instead of acting in isolation, analyzing complex wirings around a protein in a PPI network could give deeper insights into the protein’s role in the inner working of the cell than analyzing sequences of individual genes. Hence, we believe that one could lose much information by focusing on sequence information alone. We examine whether the information about homologous proteins captured by PPI network topology differs and to what extent from the information captured by their sequences. We measure how similar the topology around homologous proteins in a PPI network is and show that such proteins have statistically significantly higher network similarity than nonhomologous proteins. We compare these network similarity trends of homologous proteins with the trends in their sequence identity and find that network similarities uncover almost as much homology as sequence identities. Although none of the two methods, network topology and sequence identity, seems to capture homology information in its entirety, we demonstrate that the two might give insights into somewhat different types of biological information, as the overlap of the homology information that they uncover is relatively low. Therefore, we conclude that similarities of proteins’ topological neighborhoods in a PPI network could be used as a complementary method to sequence-based approaches for identifying homologs, as well as for analyzing evolutionary distance and functional divergence of homologous proteins.

The golden era of molecular genetic research brings about an explosion of phenotypic, genotypic and sequencing data. Building on the common aims to exploit understanding of human diseases, it also opens up an opportunity for scientific communities to share and combine research data. Genome-wide association studies (GWAS) have been widely used to locate genetic variants, which are susceptible for common diseases. In the field of medical genetics, many international collaborative consortiums have been established to conduct meta-analyses of GWAS results and to combine large genotypic data sets to perform mega genetic analyses. Having an integrated phenotype database is significant for exploiting the full potential of extensive genotypic data. In this paper, we aim to share our experience gained from integrating four heterogeneous sets of major depression phenotypic data onto the MySQL platform. These data sets constitute clinical data which had been gathered for various genetic studies for the past decade. We also highlight in this report some generic data handling techniques, the costs and benefits regarding the use of integrated phenotype database within our own institution and under the consortium framework.

Background: A chimeric transcript is a single RNA sequence which results from the transcription of two adjacent genes. Recent studies estimate that at least 4% of tandem human gene pairs may form chimeric transcripts. Affymetrix GeneChip data are used to study the expression patterns of tens of thousands of genes and the probe sequences used in these microarrays can potentially map to exotic RNA sequences such as chimeras. Results: We have studied human chimeras and investigated their expression patterns using large surveys of Affymetrix microarray data obtained from the Gene Expression Omnibus. We show that for six probe sets, a unique probe mapping to a transcript produced by one of the adjacent genes can be used to identify the expression patterns of readthrough transcripts. Furthermore, unique probes mapping to an intergenic exon present only in the MASK-BP3 chimera can be used directly to study the expression levels of this transcript. Conclusions: We have attempted to implement a new method for identifying tandem chimerism. In this analysis unambiguous probes are needed to measure run-off transcription and probes that map to intergenic exons are particularly valuable for identifying the expression of chimeras.

The identification of genes and SNPs involved in human diseases remains a challenge. Many public resources, databases and applications, collect biological data and perform annotations, increasing the global biological knowledge. The need of SNPs prioritization is emerging with the development of new high-throughput genotyping technologies, which allow to develop customized disease-oriented chips. Therefore, given a list of genes related to a specific biological process or disease as input, a crucial issue is finding the most relevant SNPs to analyse. The selection of these SNPs may rely on the relevant a-priori knowledge of biomolecular features characterising all the annotated SNPs and genes of the provided list. The bioinformatics approach described here allows to retrieve a ranked list of significant SNPs from a set of input genes, such as candidate genes associated with a specific disease. The system enriches the genes set by including other genes, associated to the original ones by ontological similarity evaluation. The proposed method relies on the integration of data from public resources in a vertical perspective (from genomics to systems biology data), the evaluation of features from biomolecular knowledge, the computation of partial scores for SNPs and finally their ranking, relying on their global score. Our approach has been implemented into a web based tool called SNPRanker, which is accessible through at the URL http://www.itb.cnr.it/snpranker. An interesting application of the presented system is the prioritisation of SNPs related to genes involved in specific pathologies, in order to produce custom arrays.

The availability of various high-throughput experimental and computational methods developed in the last decade allowed molecular biologists to investigate the functions of genes at system level opening unprecedented research opportunities. Despite the automated prediction of genes functions could be included in the most difficult problems in bioinformatics, several recently published works showed that consistent improvements in prediction performances can be obtained by integrating heterogeneous data sources. Nevertheless, very few works have been dedicated to the investigation of the impact of noisy data on the prediction performances achievable by using data integration approaches. In this contribution we investigated the tolerance of multiple classifier systems (MCS) to noisy data in gene function prediction experiments based on data integration methods. The experimental results show that performances of MCS do not undergo a significant decay when noisy data sets are added. In addition, we show that in this task MCS are competitive with kernel fusion, one of the most widely applied technique for data integration in gene function prediction problems.