We present a new approach to pattern discovery called string pattern regression, where we are given a data set that consists of a string attribute and an objective numerical attribute. The problem is to find the best string pattern that divides the data set in such a way that the distribution of the numerical attribute values of the set for which the pattern matches the string attribute, is most distinct, with respect to some appropriate measure, from the distribution of the numerical attribute values of the set for which the pattern does not match the string attribute. By.solving this problem, we are able to discover, at the same time, a subset of the data whose objective numerical attributes are significantly different from rest of the data, as well as the splitting rule in the form of a string pattern that is conserved in the subset. Although the problem can be solved in linear time for thesubstring pattern class, the problem is NP-hard in the general case (i.e. more complex patterns), and we present an exact but efficient branch-and-bound algorithm which is applicable to various pattern classes. We apply our algorithm to intron sequences of human, mouse, fly, and zebrafish, and show the practicality of our approach and algorithm. We also discuss possible extensions of our algorithm, as well as promising applications, such as microarray gene expression data.

View full abstract

With the increasing amount of available genome sequences, novel tools are needed for comprehensive analysis of species-specific sequence characteristics for a wide variety of genomes. We used an unsupervised neural network algorithm, Kohonen's self-organizing map (SOM), to analyze diand trinucleotide frequencies in 9 eukaryotic genomes of known sequences (a total of 1.2 Gb); S. cerevisiae, S. pombe, C. elegans, A. thaliana, D. melanogaster, Fugu, and rice, as well as P. falciparum chromosomes 2 and 3, and human chromosomes 14, 20, 21, and 22, that have been almost completely sequenced. Each genomic sequence with different window sizes was encoded as a 16-and 64-dimensional vector giving relative frequencies of di- and trinucleotides, respectively. From analysis of a total of 120, 000 nonoverlapping 10-kb sequences and overlapping 100-kb sequences with a moving step size of 10 kb, derived from a total of the 1.2 Gb genomic sequences, clear species-specific separations of most sequences were obtained with the SOMs. The unsupervised algorithm could recognize, in most of the 120, 000 10-kb sequences, the species-specific characteristics (key combinations of oligonucleotide frequencies) that are signature representations of each genome. Because the classification power is very high, the SOMs can provide fundamental bioinformatic strategies for extracting a wide range of genomic information that could not otherwise be obtained.

View full abstract

In the present study, we identified periodic patterns in nucleotide sequence, and characterizednucleotide sequences that confer periodicities to Arabidopsis thaliana and Drosophila melanogaster on the basis of a power spectrum method and frequency of nucleotide sequences. To assign regions that contribute to each periodicity we calculated periodic nucleotide distributions by a parameter proposed in the paper. In A. thaliana, we obtained three periodicities (248 bp-, 167 bp-, and 126 bp) in chromosome 3, three peaks (174 bp-, 88 bp-, and 59 bp-period) in chromosome 4, and four periodicities (356 bp, 174 bp, 88 bp, and 59 bp) in chromosome 5. These are relation to ORF that consists of Gly-rich amino acid sequences including histone protein that consists of Gly-, Ser-, and Ala-rich amino acids residues. For D. melanogaster genome we found that G or C spectral curves have flat region at middle frequency range from f=10^-4 to 10^-5 (corresponding to cyclic size 1 kb-5 kb), which may be associated with randomness of base sequence composition. This property has not been observed in Saccharomyces cerevisiae, Caenorhabditis elegans, and Homo sapiens yet.

View full abstract

As sequenced genomes become larger and sequencing process becomes faster, there is a need to develop a tool to analyze sequences in the whole genomic scale. However, on-memory algorithms such as suffix tree and suffix array are not applicable to the analysis of whole genome sequence set, since the size of individual whole genome ranges from several million base pairs to hundreds billion base pairs. In order to effectively manipulate the huge sequence data, it is necessary to use the indexed data structure for external memory. In this paper, we introduce a workbench called SequeX for the analysis and visualization of whole genome sequences using SSB-tree (Static SB-tree). It consists of two parts: the analysis query subsystem and the visualization subsystem. The query subsystem supports various transactions such as pattern matching, k-occurrence, and k-mer analysis. The visualization subsystem helps biologists to easily understand whole genome structure and feature by sequence viewer, annotation viewer, CGR (Chaos Game Representation) viewer, and k-mer viewer. The system also supports a user-friendly programming interface based on Java script for batch processing and the extension for a specific purpose of a user. SequeX can be used to identify conserved genes or sequences by the analysis of the common k-mers and annotation. We analyze the common k-mer for 72 microbial genomes announced by Entrez, and find an interesting biological fact that the longest common k-mer for 72 sequences is 11-mer, and only 11 such sequences exist. Finally we note that many common k-mers occur in conserved region such as CDS, rRNA, and tRNA.

View full abstract

Protein-protein interaction plays a critical role in biological processes. The identification of interacting proteins by computational methods can provide new leads in functional studies of uncharacterized proteins without performing extensive experiments. We developed a database for the potentially interacting domain pairs (PID) extracted from a dataset of experimentally identified interacting protein pairs (DIP: database of interacting proteins) with InterPro, an integrated database of protein families, domains and functional sites. In developing protein interaction databases and predictive methods, sensitive statistical scoring systems is critical to provide a reliability index for accurate functional analysis of interaction networks. We present a statistical scoring system, named “PID matrix score” as a measure of the interaction probability (interactability) between domains. This system provided a valuable tool for functional prediction of unknown proteins. For the evaluation of PID matrix, cross validation was performed with subsets of DIP data. The prediction system gives about 50% sensitivity and more than 98% specificity, which implies that the information for interacting proteins pairs could be enriched about 30 fold with the PID matrix. Itis demonstrated that mapping of the genome-wide interaction network can be achieved by using the PID matrix.

View full abstract

Feature selection plays an important role in classification. We present a comparative study on six feature selection heuristics by applying them to two sets of data. The first set of data are gene expression profiles from Acute Lymphoblastic Leukemia (ALL) patients. The second set of data are proteomic patterns from ovarian cancer patients. Based on features chosen by these methods, error rates of several classification algorithms were obtained for analysis. Our results demonstrate the importance of feature selection in accurately classifying new samples.

View full abstract

In recent years, the analysis of orthologous genes based on phylogenetic profiles has received popularity in bioinfomatics. We propose a new method to extract organism groups and their hierarchy from phylogenetic profiles using the independent component analysis (ICA). The method involves first finding independent axes in the projected space from the multivariate data matrix representing phylogenetic profiles for a number of orthologous genes. Then the extracted axes are correlated with major organism groups, according to the extent of affiliaion of axes scores for all the genes to specific organisms. The ICA was applied to the phylogenetic profiles created for 2875 orthologs in 77 organisms by using the KEGG/GENES database. The 9 extracted components out of 18 predefined components well represented the organism groups as categorized in KEGG. Furthermore, we performed the cluster analysis and obtained the hierarchy of organism groups.

View full abstract

The origin and early evolution of genetic codon system and early mRNAs were analyzed from a viewpoint of primordial gene theory and the poly-tRNA theory. A hypothetical 25-amino acid (aa)-primordial peptide was deduced from internal aa-sequence homology of adenylate kinases. Theoretical models were made which can reasonably explain how primitive tRNA (s) could have had converted to be earliest mRNAs via interactions between presumptive anticodons and (poly-) tRNA ribozyme. Transfer-RNA gene clusters in the trrnD- and rrnB-operons of Bacillus subtilis seemed to be relics of early peptide-synthesizing RNA machine. Detailed analyses revealed that the poly-tRNA regions in these operons are true relics of RNA-machine for making a 16-aa trrnD-peptide and a 21-aa rrnB-peptide, whose aa sequences are in the order of aa specificities of tRNAs in the tRNA gene clusters of the trrnD-operon and rrnB-operon, respectively. The primordial gene-encoded peptide deduced from adenylate kinases were found to be a genuine homologue of the rrnB-peptide. Various protein superfamilies were found to have evolved from either of these two types of primitive peptides. Earlist mRNAs were concluded to have evolved from tRNA^GlY (trrnD-mRNA) or tRNA^His (rrnB-mRNA), where trrnD-and rrnB-mRNAs are hypothetical primitive mRNAs complementary to the tandem arrangement of 16 or 21 anticodons of tRNAs in the trrnD-operon and rrnB-operon, respectively. The poly-tRNA model is considered to be an excellent theory, because it can reasonably explain origins of both genetic codes and earliest mRNAs, and because the hypothesis can be statistically evaluated by base-identity levels in proper alignments. The genetic codon system is a typical mature semeiotic system within a cell, and the genesis of the geneic codon system was discussed from an aspect of de Saussure's semeiology. Arbitrary correspondence between (anti) codon and aa would be most plausibly a result of semeiotic culture system of intracellular tRNA-riboorganismic society.

View full abstract

Evolutionary trees sit at the core of all realistic models describing a set of related sequences, including alignment, homology search, ancestral protein reconstruction and 2D/3D structural change. It is important to assess the stochastic error when estimating a tree, including models using the most realistic likelihood-based optimizations, yet computation times may be many days or weeks. If so, the bootstrap is computationally prohibitive. Here we show that the extremely fast “resampling of estimated log likelihoods” or RELL method behaves well under more general circumstances than previously examined. RELL approximates the bootstrap (BP) proportions of trees better that some bootstrap methods that rely on fast heuristics to search the tree space. The BIC approximation of the Bayesian posterior probability (BPP) of trees is made more accurate by including an additional term related to the determinant of the information matrix (which may also be obtained as a product of gradient or score vectors). Such estimates are shown to be very close to MCMC chain values. Our analysis of mammalian mitochondrial amino acid sequences suggest that when model breakdown occurs, as it typically does for sequences separated by more than a few million years, the BPP values are far to peaked and the real fluctuations in the likelihood of the data are many times larger than expected. Accordingly, several ways to incorporate the bootstrap and other types of direct resampling with MCMC procedures are outlined. Genes evolve by a process which involves some sites following a tree close to, but not identical with, the species tree. It is seen that under such a likelihood model BP (bootstrap proportions) and BPP estimates may still be reasonable estimates of the species tree. Since many of the methods studied are very fast computationally, there is no reason to ignore stochastic error even with the slowest ML or likelihood based methods.

View full abstract

Human DNA consists of a large number of tandem repeat sequences. Such sequences are usually called satellites, with the primary example being the centromeric alpha-satellite DNA. The basic repeat unit of the alpha-satellite DNA is a 171bp monomer. However, with the exception of peripheral alpha-satellite DNA, monomers can be grouped into blocks of k-monomers (4<k<20) between which the divergence rate is much smaller (e.g. 5%). Perhaps the simplest and best understood mechanism for tandem repeat array evolution is the unequal crossover. Although it is possible that the alpha-satellite sequence developed as a result of subsequent unequal crossovers only, no formal computational framework seems to have been developed to verify this possibility. In this paper we develop such a framework and perform experiments which seem to indicate that pericentromeric alpha-satellite segments (which are devoid of higher-order structure) are evolutionarily distinct from the higher-order repeat segments. It is likely that the higher order repeats developed independently in distinct regions of the genome and were carried into their current locations through an unknown mechanism of transposition.

View full abstract

Current algorithmic studies of genome rearrangement ignore the length of reversals (or inversions); rather, they only count their number. We introduce a new cost model in which the lengths of the reversed sequences play a role, allowing more flexibility in accounting for mutation phenomena. Our focus is on sorting unsigned (unoriented) permutations by reversals; since this problem remains difficult (NP-hard) in our new model, the best we can hope for are approximation results. We propose an efficient, novel algorithm that takes (a monotonic function f of) length into account as an optimization criterion and study its properties. Our results include an upper bound of O (f (n) lg²n) for any additive cost measure f on the cost of sorting any n-element permutation, and a guaranteed approximation ratio of O (lg²n) times optimal for sorting a given permutation. Our work poses some interesting questions to both biologists and computer scientists and suggests some new bioinformatic insights that are currently being studied.

View full abstract

We present novel kernels that measure similarity of two RNA sequences, taking account of their secondary structures. Two types of kernels are presented. One is for RNA sequences with known secondary structures, the other for those without known secondary structures. The latter employs stochastic context-free grammar (SCFG) for estimating the secondary structure. We call the latter the marginalized count kernel (MCK). We show computational experiments for MCK using 74 sets of human tRNA sequence data:(i) kernel principal component analysis (PCA) for visualizing tRNA similarities, (ii) supervised classification with support vector machines (SVMs). Both types of experiment show promising results for MCKs.

View full abstract

This paper presents a parallel hybrid genetic algorithm (GA) for solving the sum-of-pairs multiple protein sequence alignment. A new chromosome representation and its corresponding genetic operators are proposed. A multi-population GENITOR-type GA is combined with local search heuristics. It is then extended to run in parallel on a multiprocessor system for speeding up. Experimental results of benchmarks from the BAliBASE show that the proposed method is superior to MSA, OMA, and SAGA methods with regard to quality of solution and running time. It can be used for finding multiple sequence alignment as well as testing cost functions.

View full abstract

Single particle analysis is one of the methods for structural studies of protein and macromolecules developed in image analysis on electron microscopy. Reconstructing 3D structure from microscope images is not an easy analysis because of the low resolution of images and lack of the directional information of images in 3D structure. To improve the resolution, different projections are aligned, classified and averaged. Inferring the orientations of these images is so difficult that the task of reconstructing 3D structures depends upon the experience of researchers. But recently, a method to reconstruct 3D structures is automatically devised. In this paper, we propose a new method for determining Euler angles of projections by applying Genetic Algorithms (i. e., GAs).We empirically show that the proposed approach has improved the previous one in terms of computational time and acquired precision.

View full abstract

We developed maximum clique-based algorithms for spot matching for two-dimensional gel electrophoresis images, protein structure alignment and protein side-chain packing, where these problems are known to be NP-hard. Algorithms based on direct reductions to the maximum clique can find optimal solutions for instances of size (the number of points or residues) up to 50-150 using a standard PC. We also developed pre-processing techniques to reduce the sizes of graphs. Combined with some heuristics, many realistic instances can be solved approximately.

View full abstract

We examine the use of deformation propensity at individual base steps for the identification of DNA-protein binding sites. We have previously demonstrated that estimates of the total energy to bend DNA to its bound conformation can partially explain indirect DNA-protein interactions. We now show that the deformation propensities at each base step are not equally informative for classifying a sequence as a binding site, and that applying non-uniform weights to the contribution of each base step to aggregate deformation propensity can greatly improve classification accuracy. We show that a perceptron can be trained to use the deformation propensity at each step in a sequence to generate such weights.

View full abstract

Divergence in sequence through evolution precludes sequence alignment based homology methodologies for protein folding prediction from detecting structural and folding similarities for distantly related protein. Homolog coverage of actual data bases is also a factor playing a critical role in the performance of those methodologies, the factor being conspicuously apparent in what is called the twilight zone of sequence homology in which proteins of high degree of similarity in both biological function and structure are found but for which the amino acid sequence homology ranges from about 20% to less than 30%. In contrast to these methodologies a strategy is proposed here based on a different concept of sequence homology. This concept is derived from a periodicity analysis of the physicochemical properties of the residues constituting proteins primary structures. The analysis is performed using a front-end processing technique in automatic speech recognition by means of which the cepstrum (measure of the periodic wiggliness of a frequency response) is computed that leads to a spectral envelope that depicts the subtle periodicity in physicochemical characteristics of the sequence. Homology in sequences is then derived by alignment of spectral envelopes. Proteins sharing common folding patterns and biological function but low sequence homology can then be detected by the similarity in spectral dimension. The methodology applied to protein folding recognition underscores in many cases other methodologies in the twilight zone.

View full abstract

Chloroplast genome originates from the genome of ancestral cyanobacterial endosymbiont. The comparison of the genomes of cyanobacteria and plants has been made possible by the advance in genome sequencing. I report here current results of our computational efforts to compare the genomes of cyanobacteria and plants and to trace the process of evolution of cyanobacteria, chloroplasts and plants. Cyanobacteria form a clearly defined monophyletic clade with reasonable level of diversity and are ideal for testing various approaches of genome comparison. Analysis of short sequence features such as genome signature was found to be useful in characterizing cyanobacterial genomes. Comparison of genome contents was performed by homology grouping of predicted protein coding sequences, rather than orthologue-based comparison, to minimize effects of multidomain proteins and large protein families, both of which are important in cyanobacterial genomes. Comparison of the genomes of six species of cyanobacteria suggests that there are a number of species-specific additions of protein genes, and this information is useful in reconstructing phylogenetic relationship. The homology groups in cyanobacteria were used as a reference to compare plants and non-photosynthetic organisms. The results suggest that 238 groups that are common to all organisms analyzed may define a minimal set of gene groups. In addition, only 80 groups are identified as the gene groups that could not have been acquired by plants without cyanobacterial endosymbiosis. Further study is needed to identify plant genes of cyanobacterial origin.

View full abstract

We introduce here a novel index which precisely derives protein coding regions from cross-species genome alignments. The index is deeply related to frame recovery observed in coding sequence alignments, that is, if insertions or deletions of nucleotides causes frame shifts in coding regions, other in-dels which recover the reading frames will be often observed in the vicinity. In contrast, such frame recoveries are not observed in other conserved regions. We prepared two gene models: a model which finds gene by using sequence similarity and intrinsic gene measures (basic model), and the other model which finds gene by using frame recovery index in addition to sequence similarity and intrinsic gene measures (frame recovery model). We evaluated the prediction accuracies of the two models, and our benchmark test revealed that frame recovery model significantly improved the prediction accuracy in comparison with basic model.

View full abstract

Correct prediction of the translation initiation site (TIS) is an important issue in genomic research. We show that feature generation together with correlation based feature selection can be used with a variety of machine learning algorithms to give highly accurate translation initiation site prediction. Only very few features are needed and the results achieve comparable accuracy to the best existing approaches. Our approach has the advantage that it does not require one to devise a special prediction method; rather standard machine learning classifiers are shown to give very good performance on the selected features. The raw and generated features which we have found to be important are the following: positions-3 and-1 in the sequence; upstream k-grams for k=3, 4, and 5; stop-codon frequency; downstream in-frame 3-gram; and the distance of ATG to the beginning of the sequence. The best result, with an overall accuracy of 90%, is obtained by selecting only seven features from this set. The same features retrained with the use of a scanning model achieves an overall accuracy of 94% on this dataset.

View full abstract

Detailed classifications, controlled vocabularies and organised terminology are widely used in different areas of science and technology. Their relatively recent introduction in molecular biology has been crucial for progress in the analysis of genonics and massive proteomics experiments. Unfortunately the construction of the ontologies, including terminology, classification and entity relations requires considerable effort, including the analysis of massive amounts of literature. We propose here a method that automatically generates classifications of gene-product functions using bibliographic information. The corresponding classification structures mirror the ones constructed by human experts. The analysis of a large structure built for yeast gene-products, and the detailed inspection of various examples, show encouraging properties. In particular, the comparison with the well accepted GO ontology points to different situations in which the automatically derived classification can be useful for assisting human experts in the annotation of ontologies.

View full abstract

The intracellular metabolic fluxes can be calculated by metabolic flux analysis, which uses a stoichiometric model for the intracellular reactions along with mass balances around the intracellular metabolites. In this study, we have constructed in silico metabolic pathway network of Escherichia coli consisting of 301 reactions and 294 metabolites. Metabolic flux analyses were carried out to estimate flux distributions to achieve the maximum in silico yield of succinic acid in E. coli. The maximum in silico yield of succinic acid was only 83% of its theoretical yield. The lower in silico yield of succinic acid was found to be due to the insufficient reducing power, which could be increased to its theoretical yield by supplying more reducing power. Furthermore, the optimal metabolic pathways for the production of succinic acid could be proposed based on the results of metabolic flux analyses. In the case of succinic acid production, it was found that pyruvate carboxylation pathway should be used rather than phosphoenolpyruvate carboxylation pathway for its optimal production in E. coli. Then, the in silico optimal succinic acid pathway was compared with conventional succinic acid pathway through minimum set of wet experiments. The results of wet experiments indicate that the pathway predicted by in silico analysis is more efficient than conventional pathway.

View full abstract

The ultimate goal of bioinformatics is to reconstruct biological systems in a computer. Biological systems have a multi-scale and multi-level biological hierarchy. The cellular level of the hierarchy is appropriate and practicable for reconstructing biological systems by computer modeling. In our first application of computer modeling to development of the nematode C. elegans, we focus on the cellular arrangement in early embryos. This plays a very important role in cell fate determination by cell-cell interaction, which is largely restricted by physical conditions. We have already constructed a computer model of a C. elegans embryo, currently up to the 4-cell stage, using deformable and dividable geometric graphics. Modeling components of the embryo are based solely on cellular-level dynamics. Here, we modeled new physical phenomena of cell division, cell rounding and stiffening; we then combined them with already modeled phenomena, contractile ring contraction and cell elongation. We investigated effectiveness of the new model on cellular arrangement by computer simulations. We found that cell rounding and stiffening only during the period of cell division were effective to generate almost identical cellular arrangements to in real embryos. Since cells could be soft during the period between cell divisions, implementation of the new model resulted in cell shapes similar to real embryos. The nature of the model and its relationship to real embryos arediscussed.

View full abstract