The complete genomic nucleotide sequence data of more than 10 unicellular organisms have become available. During the past years, we have been focusing our attention to the analysis of the structure and function of the ribosome and its protein components. By making use of the genomic sequence data, our work can now be extended to comparative analysis of the ribosomal compo-nents at the genomic level. Such analysis will contribute to our understanding of the structure-function relationship of the ribosome that is vital to the expression of genetic information. Bearing these in mind, the ribosomal protein genes of organisms whose genomic sequence data are available were analyzed, which included Aquifex aeolicus; Archaeoglobus fulgidus; Borrelia burgdorferi; Bacillus subtilis; Escherichia coli; Haemophilus influenzae; Helicobacter pylori; Methanococcus jan-naschii; Mycoplasma genitalium; Mycoplasma pneumoniae; Synechosystis sp., and Saccharomyces cerevisiae. In addition, the amino acid sequence data of Bacillus stearothermophilus ribosomal pro-teins were used in the evolutionary evaluation. The results indicate that, in eubacteria including two species of Mycoplasma, the operon structure of ribosomal protein genes is well conserved, while their relative orientation and chromosomal location are diverged into several classes. The operon structure in M. jannaschii on the other hand is quite different from the eubacterial one and we noticed that its many genes show similarity to rat ribosomal protein genes. The degrees of sequence conservation differ from one ribosomal protein gene to another, but several genes encoding proteins that are considered to be of structural importance are conserved throughout the bacterial species including archaebacteria and further in S. cerevisiae.

View full abstract

Nucleotide sequences of small-subunit rRNA (16S rRNA) are most commonly used for the identification and characterization of bacteria and their complex communities. However, 16S rRNA evolves slowly and is often not very convenient to resolve bacterial strains at the species level. We have therefore attempted to develop a rapid and more convenient system for bacterial identification using the gyrB gene sequences. We chose the gyrB gene, because (i) it is rarely transmitted horizontally, (ii) its molecular evolution rate is higher than that of 16S rRNA, and (iii) the gene is distributed ubiquitously among bacterial species. We PCR-amplified the 1.2 kb-long gyrB segments from about 1, 000 bacterial species by using degenerate primers and determined their nucleotide sequences. The resultant data have been assembled into the gyrB database accessible via WWW.

View full abstract

The method of phylogenetic invariants was developed to apply to aligned sequence data generated, according to a stochastic substitution model, for N species related through an unknown phylogenetic tree. The invariants are functions of the probabilities of the observable N-tuples, which are identically zero, over all choices of branch length, for some trees. Evaluating the invariants associated with all possible trees, using observed N-tuple frequencies over all sequence positions, enables us to rapidly infer the generating tree.
An aspect of evolution at the genomic level much studied recently is the rearrangements of gene order along the chromosome from one species to another. Instead of the substitutions responsible for sequence evolution, we examine the non-local processes responsible for genome rearrangements such as inversion of arbitrarily long segments of chromosomes. By treating the potential adjacency of each possible pair of genes as a “position”, an appropriate “substitution” model can be recognized as governing the rearrangement process, and a probabilistically principled phylogenetic inference can be set up. We calculate the invariants for this process for N=5, and apply them to mitochondrial genome data from coelomate metazoans, showing how they resolve key aspects of branching order.

View full abstract

In order to fully make use of the vast amount of information in the complete genome sequences, we are developing a genome-scale system for predicting gene functions and cellular functions. The system makes use of the information of sequence similarity, the information of positional correlations in the genome, and the reference knowledge stored as the ortholog group tables in KEGG (Kyoto Encyclopedia of Genes and Genomes). The ortholog group table summarizes orthologous and paralogous relations among different organisms for a set of genes that are considered to form a functional unit, such as a conserved portion of the metabolic pathway or a molecular machinery for the membrane transport. At the moment, the ortholog group table is constructed for the cases where the genes are clustered in physically close positions in the genome for at least one organism. In this paper, we describe the system and the actual analysis of the complete genome of Pyrococcus horikoshii to identify ABC transporters.

View full abstract

Retrotransposons have been generally known to integrate randomly into host genomes. Jurka (1997)[3], however, showed consensus sequence patterns at integration sites of certain mammalian retrotransposons, and suggested involvement of sequence specific enzymes that mediate integration.
We have conducted comprehensive sequence analyses of 5' flanking regions of primate Alu elements. In contrast to the small but clean data set Jurka (1997)[3] used, (1) larger number of samples were used, (2) wider region of 5' end of Alu elements was analyzed, and (3) comparisons were made among different subfamilies for comprehensive analyses in order to identify characteristic sequence pattern (s) preceding 5' end of Alu elements. The nucleotide occurrences at each position within 500 bases of 5' end of Alus were counted to obtain profiles. Information content at each nucleotide position in the same region was, then, computed. Distinctive difference in the nucleotide composition and information content values that divides the region into two was observed. The region between -20 and 5' end of Alu elements is found to be highly adenine-rich and shows significantly higher information content values compared to the rest of the region, implying the existence of certain characteristic sequence pattern in this region. Also, younger subfamilies of Alu elements show higher information content values than older subfamilies. This implies that certain characteristic sequence pattern already existed in the region between -20 and 5' end of Alu elements at the time of Alu integration, and accumulation of mutation in the course of time resulted in the less distinctive sequence pattern in older sequences. Frequencies of all possible triplets (total of 64) were measured in the same region in order to identify characteristic sequence pattern (s). Observation that frequencies of triplets “aaa, ” “taa” and “tta” in the 5' flanking sequences were high is consistent with Jurka (1997)[3]. Frequencies of some other triplets such as “gaa, ” “caa, ” “aac, ” “ctt, ” “gtt, ” “atg, ” etc. which do not comprise the primary candidates for the nick site in Jurka (1997), also show significantly high frequencies.

View full abstract

The structural significance of introns in α and β globin genes were evaluated in terms of the folding (or stacking) free energy of the nascent RNA sequences as obtained from secondary structure calculations. From calculations of the folding free energy of exons and introns, it was found that introns tend to have a more folded (globular) structure compared to the exon counterparts segments. The results suggest that significant biological information is present in all intronic sequences including spliceosome type introns.

View full abstract

We have selected the most frequently seen verbs from raw texts made up of 1-million-words of Medline abstracts, and we were able to identify (or bracket) noun phrases contained in the corpus, with a precision rate of 90%. Then, based on the noun-phrase-bracketted corpus, we tried to find the subject and object terms for some frequently seen verbs in the domain. The precision rate of finding the right subject and object for each verb was about 73%. This task was only made possible because we were able to linguistically analyze (or parse) a large quantity of a raw corpus. Our approach will be useful for classifying genes and gene products and for identifying the interaction between them. It is the first step of our effort in building a genome-related thesaurus and hierarchies in a fully automatic way.

View full abstract

Gathering data on molecular interactions to be fed into a specialized database has motivated the development of a computer system to help extracting pertinent information from texts, relying on advanced linguistic tools, completed with object-oriented knowledge modeling capabilities. As a first step toward this challenging objective, a program for the identification of gene symbols and names inside sentences has been devised. The main difficulty is that these names and symbols do not appear to follow construction rules. The program is thus made up of a series of sieves of different natures, lexical, morphological and semantic, to distinguish among the words of a sentence those which can only be potential gene symbols or names. Its performance has been evaluated, in terms of coverage and precision ratios, on a corpus of texts concerning D. melanogaster for which the list of names of known genes is available for checking.

View full abstract

Huge quantities of on-line medical texts such as Medline are available, and we would hope to extract useful information from these resources, as much as possible, hopefully in an automatic way, with the aid of computer technologies. Especially, recent advances in Natural Language Processing (NLP) techniques raise new challenges and opportunities for tackling genome-related on-line text; combining NLP techniques with genome informatics extends beyond the traditional realms of either technology to a variety of emerging applications. In this paper, we explain some of our current efforts for developing various NLP-based tools for tackling genome-related on-line documents for information extraction task.

View full abstract

We consider the problem of selecting the articles of experts' interest from a literature database with the assistance of a machine learning system. For this purpose, we propose the rough reading strategy which combines the experts' knowledge with the machine learning system. For the articles converted through the rough reading strategy, we employ the learning system BONSAI and apply it for discovering rules which may reduce the work of experts in selecting the articles. Furthermore, we devise an algorithm which iterates the above procedure until almost all records of experts' interest are selected. Experimental results by using the articles from Cell show that almost all records of experts' interest are selected while reducing the works of experts drastically.

View full abstract

BLASTP gives a good overall indication of what function a protein might have. However, analysis of BLASTP reports to discover various domain features in the protein is still tedious. We address this problem by using the modern data integration system, Kleisli, 1 to bring out annotated features of BLASTP results. We further strengthen our solution by incorporating additional information from SEG, ClustalW, hmmPfam, etc. It is also noteworthy that the codes of our implementation is sufficiently short to be presented in its entirety.

View full abstract

Given a set of N sequences, the Multiple Sequence Alignment problem is to align these N sequences, possibly with gaps, that brings out the best commonality of the N sequences. MUSCA¹ is a two-stage approach to the alignment problem by identifying two relatively simpler sub-problems whose solutions are used to obtain the alignment of the sequences. We first discover motifs in the N sequences and then extract an appropriate subset of compatible motifs to obtain a “good” alignment. The motifs of interest to us are the irredundant motifs which are only polynomial in the input size. In practice, however, the number is much smaller (sub-linear). Notice that this step aids in a direct N-wise alignment, as opposed to composing the alignments from lower order (say pairwise) alignments and the solution is also independent of the order of the input sequences; hence the algorithm works very well while dealing with a large number of sequences. The second part of the problem that deals with obtaining a good alignment is solved using a graph-theoretic approach that computes an induced subgraph satisfying certain simple constraints. We reduce a version of this problem to that of solving an instance of a set covering problem, thus offer the best possible approximate solution to the problem (provided P≅NP). Our experimental results, while being preliminary, indicate that this approach is efficient, particularly on large numbers of long sequences, and, gives good alignments when tested on biological data such as DNA and protein sequences. We introduce the the notion of an alignment number K (2 ≤K≤N), a user-controlled parameter, that lends a useful flexibility to the aligning program: this additional requirement constrains the alignment to have at least K sequences agree on a character, whenever possible, in the alignment. The usefulness of the alignment number is corroborated by the users who view this as a natural constraint while dealing with a large number of sequences.

View full abstract

The alignment problem of DNA or protein sequences is very applicable and important in various fields of molecular biology. This problem can be reduced to the shortest path problem and Ikeda and Imai [4] showed that the A* algorithm works efficiently with the estimator utilizing all 2-dimensional sub-alignments.
In this paper we present new powerful estimators utilizing k≥3 dimensional sub-alignments, and propose a new bounding technique using VΔ, a set of vertices in the paths whose lengths are at most Δ longer than the shortest path. We also extend our algorithm to a recursive-estimate version. These algorithms become more efficient when the number of sequences increase, or the similarity among sequences is lower.

View full abstract

Protein information analysis is widely regarded as a key technology in drug design, macromolecular engineering, and understanding genome sequences. Because vast amount of calculations are required, further speed-up for protein information analysis is very much in demand. We have implemented the PAPIA (PArallel Protein Information Analysis) system on the RWC PC cluster IIa (“PAPIA cluster”) which consists of 64 Pentium Pro 200MHz microprocessors. The PAPIA system performs fast parallel processing for typical calculations in protein analysis, such as structure similarity search, sequence homology search and multiple sequence alignment, nearly 60 times faster th an a single processor. We have started a WWW service (http://www.rwcp.or.jp/papia/), allowing any biologist to easily submit jobs to the PAPIA system through a WWW browser. The user can experience the power of current parallel processing technology.

View full abstract

We study the problem of finding a genetic network from data obtained by multiple gene disruptions and overexpressions. We define a genetic network as a weighted graph, and analyze the computational complexity of the problem. We show that if there exists a weighted network which is consistent with given data, we can find it in polynomial time. Moreover, we also consider the optimization problem, where we try to find an optimally consistent weighted network with given data. We show that the problem is NP-hard. On the other hand, we give a polynomial-time approximation algorithm to solve it with approximation ratio 2. We report some simulation results on experiments.

View full abstract

A hot research topic in genomics is to analyze the interactions between genes by systematic gene disruptions and gene overexpressions. Based on a boolean network model without time delay, we have been investigating efficient strategies for identifying a genetic network by multiple gene disruptions and overexpressions. This paper first shows the relationship between our boolean network model without time delay and the standard synchronous boolean network model. Then we present a simulator of boolean networks without time delay for multiple gene disruptions and gene overexpressions, which includes a genetic network identifier with a graphic interface that generates instructions for experiments of gene disruptions and overexpressions.

View full abstract

We have developed the fully-automated algorithms for processing 2-D gel electrophoretograms based on RLGS (restriction landmark genomic scanning) method; one for fully-automated spot recognition from RLGS electrophoretogram and another for fully-automated pairwise matching of the spots found on such 2-D electrophoretograms. Without any human interaction, several thousands of spots on a 2-D electrophoretogram, including hidden spots found at the shoulder of large spots, can be identified correctly by applying our spot recognition algorithm, except for only a few true-negative and false-positive spots. Once the locations and intensities of the landmark spots are correctly recognized automatically, our pairwise spot matching algorithm reliably and rapidly identifies equivalent pairs of spots found on the nonlinearly distorted RLGS electrophoretograms in the fully-automatic way, i.e., the boring and annoying spot landmarking process isunnecessary. At the beginning of the spot matching process, most suitable pair of corresponding spots is searched automatically, then the other equivalent pairs of spots are identified. With our powerful image processing algorithms, it is possible to detect DNA molecular changes such as deletions, additions, amplifications or DNA methylations occurring at or near to the restriction enzyme cleavage sites by means of comparing large amount of RLGS electrophoretograms, without any visualinspection and human interaction.

View full abstract

We have two objectives. First, we want to build a system for detecting tetratricopeptide repeats in protein sequences. Second, we want to demonstrate how the general bioinformatics database integration system called Kleisli can help build such a system easily. We achieve these two objectives by showing that short and clear programs can be written in Kleisli, using its high-level query language CPL, to build a TPR domain hunter by integrating WU-BLAST2.0, HMMER, Entrez, and PFAM.

View full abstract

We present a strategy for generating a multiple alignment from a hidden Markov model (HMM) for low identity, low similarity protein sequences. In this approach the ordered-series-of-motifs and the motif-intervening-regions are independently modeled. We also provide a measure of multiple alignment “goodness” called the stability function to compared one alignment to another. This strategy provides a more robust HMM representing highly divergent sequence data.

View full abstract

The conditional probability, P (s|x), is a statement of the probability that the event, s, will occur given prior knowledge for the value of x. If x is given and if s is randomly distributed, then an empirical approximation of the true conditional probability can be computed by the application of Bayes' Theorem. Here s represents one of two structural classes, either ordered, s_o, or disordered, s_d, and x represents an attribute value calculated over a window of 21 amino-acids. Plots of P (slx) versus x provide information about the correlation between the given sequence attribute and disorder or order. These conditional probability plots allow quantitative comparisons between individual attributes for their ability to discriminate between order and disorder states. Using such quantitative comparisons, 38 different sequence attributes have been rank-ordered. Attributes based on cysteine, the aromatics, flexible tendencies, and charge were found to be the best attributes for distinguishing order and disorder among those tested so far.

View full abstract

Using ordered and disordered regions identified either by X-ray crystallography or by NMR spectroscopy, we trained neural networks to predict order and disorder from amino acid sequence. Although the NMR-based predictor initially appeared to be much better than the one based on the X-ray data, both predictors yielded similar overall accuracies when tested on each other's training sets, and indicated similar regions of disorder upon each sequence. The predictors trained with X-ray data showed similar results for a 5-cross validation experiment and for the out-of-sample predictions on the NMR characterized data. In contrast, the predictor trained with NMR data gave substantially worse accuracies on the out-of-sample X-ray data as compared to the accuracies displayed by the 5-cross validation during the network training. Overall, the results from the two predictors suggest that disordered regions comprise a sequence-dependant category distinct from that of ordered protein structure.

View full abstract

These multivariate time series analysis methods are supposed to be effective in analyzing metabolic network, particularly when feedback loops are present.
However, it is nearly impossible for laboratory experiments to obtain a large collection of sample values enough to employ these analytical methods. Since computer simulation can produce ample data sufficient for time series analyses, we believe that the E-CELL system will be a promising means for investigation of dynamic behavior of metabolic network.

View full abstract

Since a number genes detected were previously reported to be virulence genes, and many other candidates may be associated with virulence based on their predicted function, there is suggestive evidence that the system has the potential to identify many virulence genes from the raw genomic data of a bacterial pathogen.

View full abstract