MOTIVATION: Proteins can be naturally classified into families of homologous sequences that derive from a common ancestor. The comparison of homologous sequences and the analysis of their phylogenetic relationships provide useful information regarding the function and evolution of genes. One important difficulty of clustering methods is to distinguish highly divergent homologous sequences from sequences that only share partial homology due to evolution by protein domain rearrangements. Existing clustering methods require parameters that have to be set a priori. Given the variability in the evolution pattern among proteins, these parameters cannot be optimal for all gene families. RESULTS: We propose a strategy that aims at clustering sequences homologous over their entire length, and that takes into account the pattern of substitution specific to each gene family. Sequences are first all compared with each other and clustered into pre-families, based on pairwise similarity criteria, with permissive parameters to optimize sensitivity. Pre-families are then divided into homogeneous clusters, based on the topology of the similarity network. Finally, clusters are progressively merged into families, for which we compute multiple alignments, and we use a model selection technique to find the optimal tradeoff between the number of families and multiple alignment likelihood. To evaluate this method, called HiFiX, we analyzed simulated sequences and manually curated datasets. These tests showed that HiFiX is the only method robust to both sequence divergence and domain rearrangements. HiFiX is fast enough to be used on very large datasets. AVAILABILITY AND IMPLEMENTATION: The Python software HiFiX is freely available at http://lbbe.univ-lyon1.fr/hifix.

MOTIVATION: High-throughput measurement techniques for metabolism and gene expression provide a wealth of information for the identification of metabolic network models. Yet, missing observations scattered over the dataset restrict the number of effectively available datapoints and make classical regression techniques inaccurate or inapplicable. Thorough exploitation of the data by identification techniques that explicitly cope with missing observations is therefore of major importance. RESULTS: We develop a maximum-likelihood approach for the estimation of unknown parameters of metabolic network models that relies on the integration of statistical priors to compensate for the missing data. In the context of the linlog metabolic modeling framework, we implement the identification method by an Expectation-Maximization (EM) algorithm and by a simpler direct numerical optimization method. We evaluate performance of our methods by comparison to existing approaches, and show that our EM method provides the best results over a variety of simulated scenarios. We then apply the EM algorithm to a real problem, the identification of a model for the Escherichia coli central carbon metabolism, based on challenging experimental data from the literature. This leads to promising results and allows us to highlight critical identification issues.

Gene regulatory networks consist of direct interactions but also include indirect interactions mediated by metabolites and signaling molecules. We describe how these indirect interactions can be derived from a model of the underlying biochemical reaction network, using weak time-scale assumptions in combination with sensitivity criteria from metabolic control analysis. We apply this approach to a model of the carbon assimilation network in Escherichia coli. Our results show that the derived gene regulatory network is densely connected, contrary to what is usually assumed. Moreover, the network is largely sign-determined, meaning that the signs of the indirect interactions are fixed by the flux directions of biochemical reactions, independently of specific parameter values and rate laws. An inversion of the fluxes following a change in growth conditions may affect the signs of the indirect interactions though. This leads to a feedback structure that is at the same time robust to changes in the kinetic properties of enzymes and that has the flexibility to accommodate radical changes in the environment.

The resolution of combinatorial assortments of protein sequences into domains is a prerequisite for protein sequence interpretation. However the recognition and clustering of homologous domains from sequence databases typically scales quadratically with respect to their size which grows exponentially, making it essential to parallelize these complex bioinformatics applications. Here we demonstrate the parallelization of MKDOM2, the sequential program that has been instrumental in the construction of the PRODOM database of protein domain families. This was challenging because of (1) dependencies between program iterations, (2) their extremely heterogeneous run times and (3) communication bottlenecks that could arise because of the large size of the data. A large scale test of the new program, MPI_MKDOM2, demonstrated its robustness against heterogeneous run times, preparing the grounds for future releases of PRODOM that would otherwise be out of reach with MKDOM2 by several orders of magnitude.

The exponential increase in the submission of nucleotide sequences to the nucleotide sequence database by genome sequencing centres has resulted in a need for rapid, automatic methods for classification of the resulting protein sequences. There are several signature and sequence cluster-based methods for protein classification, each resource having distinct areas of optimum application owing to the differences in the underlying analysis methods. In recognition of this, InterPro was developed as an integrated documentation resource for protein families, domains and functional sites, to rationalise the complementary efforts of the individual protein signature database projects. The member databases - PRINTS, PROSITE, Pfam, ProDom, SMART and TIGRFAMs - form the InterPro core. Related signatures from each member database are unified into single InterPro entries. Each InterPro entry includes a unique accession number, functional descriptions and literature references, and links are made back to the relevant member database(s). Release 4.0 of InterPro (November 2001) contains 4,691 entries, representing 3,532 families, 1,068 domains, 74 repeats and 15 sites of post-translational modification (PTMs) encoded by different regular expressions, profiles, fingerprints and hidden Markov models (HMMs). Each InterPro entry lists all the matches against SWISS-PROT and TrEMBL (2,141,621 InterPro hits from 586,124 SWISS-PROT and TrEMBL protein sequences). The database is freely accessible for text- and sequence-based searches.

The ProDom database is a comprehensive set of protein domain families automatically generated from the SWISS-PROT and TrEMBL sequence databases. An associated database, ProDom-CG, has been derived as a restriction of ProDom to completely sequenced genomes. The ProDom construction method is based on iterative PSI-BLAST searches and multiple alignments are generated for each domain family. The ProDom web server provides the user with a set of tools to visualise multiple alignments, phylogenetic trees and domain architectures of proteins, as well as a BLAST-based server to analyse new sequences for homologous domains. The comprehensive nature of ProDom makes it particularly useful to help sustain the growth of InterPro.

MOTIVATION: InterPro is a new integrated documentation resource for protein families, domains and functional sites, developed initially as a means of rationalising the complementary efforts of the PROSITE, PRINTS, Pfam and ProDom database projects. RESULTS: Merged annotations from PRINTS, PROSITE and Pfam form the InterPro core. Each combined InterPro entry includes functional descriptions and literature references, and links are made back to the relevant parent database(s), allowing users to see at a glance whether a particular family or domain has associated patterns, profiles, fingerprints, etc. Merged and individual entries (i.e. those that have no counterpart in the companion resources) are assigned unique accession numbers. Release 1.2 of InterPro (June 2000) contains over 3000 entries, representing families, domains, repeats and sites of post-translational modification (PTMs) encoded by 6581 different regular expressions, profiles, fingerprints and Hidden Markov Models (HMMs). Each InterPro entry lists all the matches against SWISS-PROT and TrEMBL (more than 1000000 hits from 264333 different proteins out of 384572 in SWISS-PROT and TrEMBL).

Abstract Motivation: Multiple alignments of protein sequences are the basis of structural and functional analysis of protein families. It is however difficult even for an expert biologist to comprehend an alignment of more than 50 to 100 homologous sequences. Results: This paper presents a browser for the analysis of multiple alignments of large numbers of protein sequences. Phylogenetic trees and consensus sequences are computed and used to summarise the alignments; these data are stored in a structure called Rich Family Description. Summary alignments and trees are displayed in HTML pages and can be developed or reduced by the user. This browser is used to display the ProDom domain families on the Web. Its zooming facilities allow extracting information from alignments of more than 1000 homologous sequences. Availability: Programs are available free of charge on request from the authors. The ProDom database can be consulted at http://www.toulouse.inra.fr/prodom.html Contact: [email protected]

MOTIVATION: To extract the maximum possible information from a set of protein sequences, its modular organization must be known and clearly displayed. This is important both for structural and functional analysis. RESULTS: This paper presents an algorithm and a graphical interface called XDOM which performs a systematic analysis of the modular organization of any set of protein sequences. The algorithm is an automatic method to identify putative domains from sequence comparisons. The graphical tool displays the proteins as a set of linked boxes, corresponding to its domains. The method has been tested on a family of bacterial proteins and on whole genomes. It is currently applied to the complete SWISS-PROT database to build the PRODOM database. AVAILABILITY: XDOM is available free of charge by anonymous ftp:¿¿ftp://ftp.toulouse.inra.fr/pub/xdom¿ ¿. The ProDom database can be consulted at ¿¿http://protein.toulouse.inra.fr/prodom.html¿¿.

It is shown that there is a general framework, based on information theory, underlying many currently popular distortion measures used for speech recognition. Within this framework, three general categories of information-theoretic distortion measures are introduced: the generalized Kolmogorov variational distance, the f-divergence, and the Chernoff distance. There are two major results of this investigation. First, it is found that most of the important distortion measures used by workers in speech recognition fall out as a special case of one or another of the classes of probability-distribution dissimilarity measures. Second, the information-theoretic perspective adopted makes it possible to discover new distortion measures which may display superior speech recognition performance; one measure, the clamped log (cos beta ) distance, has been investigated experimentally, with promising results.>

Moving a speech recognition algorithm out of the laboratory and into a real-world setting introduces many variables that have the potential to degrade performance significantly. Specifically, algorithms that explicitly or implicitly measure the difference between the power spectra of frames of an unknown utterance and a reference template will be negatively affected by sources of spectral distortion unrelated to linguistic distinctions. We have performed a series of experiments aimed at a deeper understanding of several of the variables that come into play when implementing one of the best-performing algorithms, that of Itakura [1], in a particularly harsh real-world environment, that of the telephone network.