One important purpose of conducting gene expression experiments is to understand the correlation of gene expression profiles to disease states. Based on the notion of emerging patterns and an entropy-oriented discretization method, we discover groups of genes that are correlated to disease states in a significant way. In each group, every member gene constrained by a specific expression interval, unanimously occurs only in one type of cells with a maximally large frequency, but never unanimously happens in the other types of cells. According to our studies on the colon tumor dataset, such gene groups (also called patterns) can reach a frequency of 90%, providing good insight into the correlation of gene expression profiles to disease states. The patterns can be used to correctly predict whether a new cell is normal or cancerous.

View full abstract

Recent advances in biotechnology offer the ability to measure the levels of expression of thousands of genes in parallel. Analysis of such data can provide understanding and insight into gene function and regulatory mechanisms. Several machine learning approaches have been used to aid to understand the functions of genes. However, these tasks are made more difficult due to the noisy nature of array data and the overwhelming number of gene features. In this paper, we use the parallel genetic algorithm to filter out the informative genes relative to classification. By combing with the classification method proposed by Golub et al. [10] and Slonim et al. [17], we classify the data sets with tissues of different classes, and the preliminary results are presented in this paper.

View full abstract

This paper describes a new framework for microarray gene-expression data clustering. The foundation of this framework is a minimum spanning tree (MST) representation of a set of multidimensional gene expression data. A key property of this representation is that each cluster of the expression data corresponds to one subtree of the MST, which rigorously converts a multidimensional clustering problem to a tree partitioning problem. We have demonstrated that though the inter-data relationship is greatly simplified in the MST representation, no essential information is lost for the purpose of clustering. Two key advantages in representing a set of multi-dimensional data as an MST are: (1) the simple structure of a tree facilitates efficient implementations of rigorous clustering algorithms, which otherwise are highly computationally challenging; and (2) as an MSTbased clustering does not depend on detailed geometric shape of a cluster, it can overcome many of the problems faced by classical clustering algorithms. Based on the MST representation, we have developed a number of rigorous and efficient clustering algorithms, including two with guaranteed global optimality. We have implemented these algorithms as a computer software EXCAVATOR. To demonstrate its effectiveness, we have tested it on two data sets, i. e., expression data from yeast Saccharomyces cerevisiae, and Arabidopsis expression data in response to chitin elicitation.

View full abstract

A pair of distinct proteins in one organism may most closely match different parts of the same protein in another organism. A comparison of all proteins from the genome of Saccharomyces cerevisiae and all proteins from 24 prokaryotic genomes yields 1010 pairs of yeast proteins whose homologs are parts of one protein from a prokaryotic genome. Marcotte et al. [12] showed that proteins related in this manner are more likely to interact than proteins chosen at random. In this paper, we investigated whether genes coding for such proteins are also likely to be concurrently transcribed. We identified 1010 fused pairs of proteins encoded in the yeast genome and analyzed expression of the corresponding genes at the transcriptional level. We found that the transcriptional profiles of fused gene pairs are significantly closer than those of randomly selected pairs. This finding is reproducible and established by multiple distance metrics. Moreover, such pairs frequently share additional biologically relevant properties. Thus, while protein fusion patterns are not predictive of co-expression, they are an important element in explaining co-expression. This justifies the use of curated protein fusion events to help characterize gene co-expression clusters.

View full abstract

This paper presents a new method to extract a set of correlated genes with respect to multiple biological features. Relationships among genes on a specific feature are encoded as a graph structure whose nodes correspond to genes. For example, the genome is a graph representing positional correlations of genes on the chromosome, the pathway is a graph representing functional correlations of gene products, and the expression profile is a graph representing gene expression similarities. When a set of genes are localized in a single graph, such as a gene cluster on the chromosome, an enzyme cluster in the metabolic pathway, or a set of coexpressed genes in the microarray gene expression profile, this may suggest a functional link among those genes. The functional link would become stronger when the clusters are correlated; namely, when a set of corresponding genes form clusters in multiple graphs. The newly introduced heuristic algorithm extracts such correlated gene clusters as isomorphic subgraphs in multiple graphs by using inter-graph links that are defined based on biological relevance. Using the method, we found E. coli correlated gene clusters in which genes are related with respect to the positions in the genome and the metabolic pathway, as well as the 3D structural similarity. We also analyzed protein-protein interaction data by two-hybrid experiments and gene coexpression data by microarrays in S. cerevisiae, and estimated the possibility of utilizing our method for screening the datasets that are likely to contain many false positive relations.

View full abstract

Genomic Object Net is a software tool for modeling and simulating biopathways which employs the notion of hybrid functional net as its basic architechture. This paper shows how to integrate this basic architecture with XML documents for biopathway representations, simulations, and visualizations for creating a tailor-made simulation environment.

View full abstract

It is quite important to estimate the metabolic flux distribution (MFD) vectors in vivo, and to investigate the effect of culture environments on the flux distributions to uncover the metabolic regulation mechanism of microbial cells. The conventional approach is to compute the MFD using the stoichiometric equations and the measured specific rates (input and output variables). However, this method cannot give the MFD for the complex metabolic network which includes cyclic pathways. In the present investigation, we considered the method of analysing the metabolic fluxes based on ¹³C tracer experiments. In particular, we compared the different techniques of estimating the bidirectional fluxes in the metabolic networks, studying their applicability with respect to the different types of data formats obtained through GC-MS (gas chromatography-mass spectrometry) and NMR (nuclear magnetic resonance) measurements in labeling experiments. It was found that some techniques cannot be applied for GC-MS and NMR data. In the present research, therefore, a new preprocessing method for MS and NMR data was developed, to solve some of the problems encountered in the conventional approaches.

View full abstract

Many practical biological problems involve an intractable (NP-hard) search through a large space of possibilities. This paper describes preliminary results from a multi-queue variant of branchand- bound search that combines anytime and optimal search behavior. The algorithm applies to problems whose solutions may be described by an N-dimensional vector. It produces an approximate solution quickly, then iteratively improves the result over time until a global optimum is produced. A global optimum may be produced before producing its proof of global optimality. Local minima are never revisited. We describe preliminary applications to ab initio protein backbone prediction, small drug-like molecule conformations, and protein-DNA binding motif discovery. The results are encouraging, although still quite preliminary.

View full abstract

This paper presents a new method to find motifs from multiple protein sequences and multiple protein structures. The method consists of two parts: quantification and local multiple alignment. the former part, protein sequences and protein structures are transformed into sequences of real numbers and real vectors respectively. In the latter part, fixed length regions having similar shapes are located. A Gibbs sampling algorithm for sequences of real numbers/vectors is newly developed for finding common regions. The results of the comparison with a standard Gibbs sampling program show that the method is particularly useful when structural information is available.

View full abstract

A graph-based clustering method is proposed to cluster protein sequences into families, which automatically improves clusters of the conventional single linkage clustering method. Our approach formulates sequence clustering problem as a kind of graph partitioning problem in a weighted linkage graph, which vertices correspond to sequences, edges correspond to higher similarities than given threshold and are weighted by their similarities. The effectiveness of our method is shown in comparison with InterPro families in all mouse proteins in SWISS-PROT. The result clusters match to InterPro families much better than the single linkage clustering method. 77% of proteins in InterPro families are classified into appropriate clusters.

View full abstract

Subcellular localization is important for proteins to function. For the prediction of subcellular localizations, we have developed a method, SortPred, using the amino acid composition and order. The composition represents the global features, e.g., the amino acid composition in the full or partial sequences, while the order represents the local features, e.g., the amino acid sequence order. The former was represented by neural networks and the latter was represented by a hidden Markov model. This method predicted the signal peptides (SP), the mitochondrial targeting peptides (mTP), the chloroplast transit peptides (cTP), and the nuclear or cytosolic sequences (other) comparing together the previous methods, this method achieved slightly higher prediction accuracy, 86% for plant and 91% for non-plant. We analyzed the trained neural networks and hidden Markov models and found out that these models well represent the biological features of the sequences.

View full abstract

In the present work we evaluate the performance of an algorithm for the automatic recognition of binding sites in proteins as well as in other macromolecules whose interactions are involved in many cellular and physiological processes. The algorithm is a combination of an unsupervised learning algorithm-based on Kohonen self organizing maps-to characterize the properties of patches of protein solvent accessible surfaces and a filtering algorithm to establish both the physical boundaries of the patches as well as the level of contribution of different and distant atoms involved in the interaction. We have found that the algorithm performs extremely well in a set of randomly selected protein complexes for which the interaction interfaces are extracted and compared with the results of the algorithm. A statistical evaluation of the algorithm is additionally performed by analysis of the degree of hydrophobicity and hydrophilicity of the output patches and comparison with that of the observed interface constituent amino acids.

View full abstract

Relevant information about protein interactions is stored in textual sources. This sources are commonly used not only as archives of what is already known but also as information for generating new knowledge, particularly to pose hypothesis about new possible interactions that can be inferred from the existing ones. This task is the more creative part of scientific work in experimental systems. We present a large-scale analysis for the prediction of new interactions based on the interaction network for the ones already known and detected automatically in the literature.
During the last few years it has became clear that part of the information about protein interactions could be extracted with automatic tools, even if these tools are still far from perfect and key problems such as detection of protein names are not completely solved. We have developed a integrated automatic approach, called SUISEKI (System for Information Extraction on Interactions), able to extract protein interactions from collections of Medline abstracts.
Previous experiments with the system have shown that it is able to extract almost 70% of the interactions present in relatively large text corpus, with an accuracy of approximately 80% (for the best defined interactions) that makes the system usable in real scenarios, both at the level of extraction of protein names and at the level of extracting interaction between them.
With the analysis of the interaction map of Saccharomyces cerevisiae we show that interactions published in the years 2000/2001 frequently correspond to proteins or genes that were already very close in the interaction network deduced from the literature published before these years and that they are often connected to the same proteins. That is, discoveries are commonly done among highly connected entities. Some biologically relevant examples illustrate how interactions described in the year 2000 could have been proposed as reasonable working hypothesis with the information previously available in the automatically extracted network of interactions.

View full abstract

A functional analysis of protein fold interaction suggested that structural fold familieshave gradually acquired more diverse interacting partners while maintaining central biochemical interactions and functions. This means thatthe protein interaction network (map) maintains its robust architecture due to the functional constraints associated with the interactions.

View full abstract

A major effort is being undertaken to sequence an array of mammalian genomes. Coincidentally, the evolutionary relationships of the 18 presently recognized orders of placental mammals are only just being resolved. In this work we construct and analyse the largest alignments of amino acid sequence data to date. Our findings allow us to set up a series of superordinal groups (clades) to act as prior hypotheses for further testing. Important findings include strong evidence for a clade of Euarchonta+Glires (=Supraprimates) comprised of primates, flying lemurs, tree shrews, lagomorphs and rodents. In addition, there is good evidence for a clade of all placental mammals except Xenarthra and Afrotheria (=Boreotheria) and for the previously recognised clades Laurasiatheria, Scrotifera, Fereuungulata, Ferae, Afrotheria, Euarchonta, Glires, and Eulipotyphla. Accordingly, a revised classification of the placental mammals is put forward. Using this and molecular divergence-time methods, the ages of the superordinal splits are estimated. While results are strongly consistent with the earliest superordinal divergences all being gt; 65 mybp (Cretaceous period), they suffer from greater uncertainty than presently appreciated. The early primate split of tarsiers from the anthropoid lineage at '55 mybp is seen to be an especially informative fossil calibration point. A statistical framework for testing clades using SINE data is presented and reveals significant support for the tarsier/anthropoid clade, as well as the clades Cetruminantia and Whippomorpha. Results also underline our thesis that while sequence analysis can help set up hypothesised clades, SINEs obtainable from sequencing 1-2 MB regions of placental genomes are essential to testing them. In contrast, derivations suggest that empirical Bayesian methods for sequence data may not be robust estimators of clades. Our findings, including the study of genes such as TP53, make a good case for the tree shrew as a closer relative of primates than rodents, while also showing a slower rate of evolution in key cell cycle genes. Tree shrews are consequently high value experimental animals and a strong candidate for a genome sequencing initiative.

View full abstract

It has been claimed that complete genome sequences would clarify phylogenetic relationships between organisms but, up to now, no satisfying approach has been proposed to use efficiently these data. For instance, if the coding of presence or absence of genes in complete genomes gives interesting results, it does not take into account the phylogenetic information contained in sequences and ignores hidden paralogy by using a similarity-based definition of orthology. Also, concatenation of sequences of different genes takes hardly in consideration the specific evolutionary rate of each gene. At last, building a consensus tree is strongly limited by the low number of genes shared among all organisms. Here, we use a new method based on supertree construction, which permits to cumulate in one supertree the information and statistical support of hundreds of trees from orthologous gene families and to build the phylogeny of 33 prokaryotes and four eukaryotes with completely sequenced genomes. This approach gives a robust supertree, which demonstrates that a phylogeny of prokaryotic species is conceivable and challenges the hypothesis of a thermophilic origin of bacteria and present-day life. The results are compatible with the hypothesis of a core of genes for which lateral transfers are rare but they raise doubts on the widely admitted “complexity hypothesis” which predicts that this core is mainly implicated in informational processes.

View full abstract

We consider the problem of separating two distinct classes of k similar sequences of length n over an alphabet of size s that have been optimally multi-aligned. An objective function based on minimizing the consensus score of the separated halves is introduced and we present an O (k³n) heuristic algorithm and two optimal branch-and-bound algorithms for the problem. The branchand-bound algorithms involve progressively more powerful lower bound functions for pruning the O (2^k) search tree. The simpler lower bound takes O (n) time to evaluate given O (sn) global data structures and the stronger bound takes O ((k+s) n) time by virtue of an efficient solution to the problem of finding the second-maximum envelope of a set of piece-wise affine curves. In a series of empirical trials we establish the degree to which classes can be separated using our metric and the effective pruning efficiency of the two branch-and-bound algorithms.

View full abstract

Because of the increase in the size of genome sequence databases, the importance of indexing the sequences for fast queries grows. Suffix trees and suffix arrays are used for simple queries. However these are not suitable for complicated queries from huge amount of sequences because the indices are stored in disk which has slow access speed. We propose storing the indices in memory in a compressed form. We use the compressed suffix array. It compactly stores the suffix array at the cost of theoretically a small slowdown in access speed. We experimentally show that the overhead of using the compressed suffix array is reasonable in practice. We also propose an approximate string matching algorithm which is suitable for the compressed suffix array. Furthermore, we have constructed the compressed suffix array of the whole human genome. Because its size is about 2G bytes, a workstation can handle the search index for the whole data in main memory, which will accelerate the speed of solving various problems in genome informatics.

View full abstract

When a set of coregulated genes share a common structural RNA motif, e. g. a hairpin, most motif search approaches fail to locate the covarying but structurally conserved motif. There do exist methods that can locate structural RNA motifs, like FOLDALIGN, but the main problem with these methods is that they are computationally expensive. In FOLDALIGN, a major contribution to this is the use of a greedy algorithm to construct the multiple alignment. To ensure good quality many redundant computations must be made. However, by applying the greedy algorithm on a carefully selected subset of sequences, near full greedy quality can be obtained. The basic idea is to estimate the order in which the sequences entered a good greedy alignment. If such a ranking, found from all pairwise alignments, is in good agreement with the order of appearance in the multiple alignment, the core structural motif can be found by performing the greedy algorithm on just the top sequences in the ranking. The ranking used in this mini-greedy algorithm is found by using two complementing approaches: 1) When interpreting the FOLDALIGN score as an inner product (kernel), the sequences can be ranked according to their distance to their center of mass; 2) We construct an algorithm that attempts to find the K closest sequences in the vector space associated with the inner product, and the remaining sequences can be ranked by their minimum distance to any of the sequences, or to the center of mass in this set. The two approaches are compared and merged, and the results discussed. We also show that structural alignments of near full greedy quality can found in significantly reduced time, using these methods. The algorithm is being included in the SLASH (Stem-Loop Align SearcH) server available at http://www.bioinf.au.dk/slash.

View full abstract

Single nucleotide polymorphisms (SNPs) are the most frequent form of DNA variation and disease-causing mutations in many genes. Due to their abundance and slow mutation rate within generations, they are thought to be the next generation of genetic markers that can be used in a myriad of important biological, genetic, pharmacological, and medical applications [13, 3, 19, 18, 16, 14]. There are several strategies both experimental, and in-silico for SNP discovery and mapping. Experimental SNP discovery consists of a number of labourious steps that make this process complex and expensive. In-silico discovery has been proposed as an alternative discovery method that makes use and takes advantage of large data sets with potential SNP information that have been generated with other purposes and have not been used as a SNP information source yet. However, in order to successfully apply the in-silico method to large data sets, the following challenges need to be addressed: First it is necessary to build an integrated SNP pipeline that handles data processing steps smoothly from the beginning (collecting sequence information) to end (SNPs in the database). Also, SNP detection tool parameters have to be optimized to satisfy specific goals of the project. Finally, SNP data could not be fully used until the in-silico method is validated experimentally. In this paper we present a design and implementation of an in-silico SNP detection software pipeline that exploits the existence of large EST (expressed sequence tag) data sets and effectively addresses the above challenges. First, the pipeline allows for smooth data transition between its different components by implementing data interfaces that translate the data formats of the different tools in the different stages. Second, we optimized PolyBayes parameters for SNP detection in maize EST. Finally, we implemented a user interface that along with the database structure created allows the scientist to perform preliminary analysis of the data and to perform basic statistics on the SNP data prior to experimental validation. The pipeline works with two different types of sequence assemblers (PHRAP [20] and CAT from DoubleTwist [21]). It uses a Bayesian engine for SNP detection (PolyBayes), selects relevant polymorphism information which is then uploaded into a database. We detected 2439 SNPs and 822 insertion deletions (INDELs) with a PolyBayes probability higher than 0.99 on the public set of 68, 000 maize ESTs. The user interface allowed us analyzing the polymorphism information right after discovery in several ways that allowed us to gain insight into the distribution and significance of the newly acquired data.

View full abstract

The rapid explosion in the amount of biological data being generated worldwide is surpassing efforts to manage analysis of the data. As part of an ongoing project to automate and manage bioinformatics analysis, the authors have designed and implemented a simple automated annotation system, which is described in this paper. The system is applied to existing GenBank/DDBJ/EMBL entries and compared with existing annotations to illustrate not only potential errors but also that they are generally not up-to-date, as a result of new versions of analysis tools and updates of genomic repositories. We highlight the important Bioinformatics issues of storage and management of information to ensure data and results are kept up-to-date in light of new information becoming available. Surprisingly, from just four database entries, a significant number of new features were found. We describe the results as well as identify important issues that need to be addressed in order to automate the re-analysis/re-annotation of genomic sequences within a reasonable timeframe.

View full abstract

DNA methylation occurring within the context of CpG site in the promoter element generally correlates with transcriptional silencing, delayed replication, condensed chromatin and mammalian genome imprinting [2]. We have been focusing attention on Restriction Landmark Genomic Scanning (RLGS) method as the powerful experimental procedure which can used to determine DNA methylation status change occurring on a genome simultaneously. To support systematic analysis with RLGS method, we have developed the automated processing algorithms for two-dimensional electrophoretograms of genomic DNA based on RLGS method (RLGS profile). Our powerful processing algorithms realize the automated spot recognition from RLGS profiles and the automated comparison of a huge number of such profiles. To make systematic RLGS analysis more easy and objective, we equipped our automated RLGS image processing algorithms with a WWW interface and relational database which stores the RLGS profile images together with their accompanying information, such as DNA sample specification, two-dimensional electrophoresis conditions and so forth. The web based RLGS image processing system “DNAinsight” can be accessed via URL http://www.methylome.jp/DNAinsight/.

View full abstract