Modeling High-Dimensional Audio Sequences with Recurrent Neural Networks

Cette thèse étudie des modèles de séquences de haute dimension basés sur des réseaux de neurones récurrents (RNN) et leur application à la musique et à la parole. Bien qu'en principe les RNN puissent représenter les dépendances à long terme et la dynamique temporelle complexe propres aux séquences d'intérêt comme la vidéo, l'audio et la langue naturelle, ceux-ci n'ont pas été utilisés à leur plein potentiel depuis leur introduction par Rumelhart et al. (1986a) en raison de la difficulté de les entraîner efficacement par descente de gradient. Récemment, l'application fructueuse de l'optimisation Hessian-free et d'autres techniques d'entraînement avancées ont entraîné la recrudescence de leur utilisation dans plusieurs systèmes de l'état de l'art. Le travail de cette thèse prend part à ce développement. L'idée centrale consiste à exploiter la flexibilité des RNN pour apprendre une description probabiliste de séquences de symboles, c'est-à-dire une information de haut niveau associée aux signaux observés, qui en retour pourra servir d'à priori pour améliorer la précision de la recherche d'information. Par exemple, en modélisant l'évolution de groupes de notes dans la musique polyphonique, d'accords dans une progression harmonique, de phonèmes dans un énoncé oral ou encore de sources individuelles dans un mélange audio, nous pouvons améliorer significativement les méthodes de transcription polyphonique, de reconnaissance d'accords, de reconnaissance de la parole et de séparation de sources audio respectivement. L'application pratique de nos modèles à ces tâches est détaillée dans les quatre derniers articles présentés dans cette thèse. Dans le premier article, nous remplaçons la couche de sortie d'un RNN par des machines de Boltzmann restreintes conditionnelles pour décrire des distributions de sortie multimodales beaucoup plus riches. Dans le deuxième article, nous évaluons et proposons des méthodes avancées pour entraîner les RNN. Dans les quatre derniers articles, nous examinons différentes façons de combiner nos modèles symboliques à des réseaux profonds et à la factorisation matricielle non-négative, notamment par des produits d'experts, des architectures entrée/sortie et des cadres génératifs généralisant les modèles de Markov cachés. Nous proposons et analysons également des méthodes d'inférence efficaces pour ces modèles, telles la recherche vorace chronologique, la recherche en faisceau à haute dimension, la recherche en faisceau élagué et la descente de gradient. Finalement, nous abordons les questions de l'étiquette biaisée, du maître imposant, du lissage temporel, de la régularisation et du pré-entraînement.

A Monte Carlo method for accelerating the computation of animated radiosity sequences

Realistic rendering animation is known to be an expensive processing task when physically-based global illumination methods are used in order to improve illumination details. This paper presents an acceleration technique to compute animations in radiosity environments. The technique is based on an interpolated approach that exploits temporal coherence in radiosity. A fast global Monte Carlo pre-processing step is introduced to the whole computation of the animated sequence to select important frames. These are fully computed and used as a base for the interpolation of all the sequence. The approach is completely view-independent. Once the illumination is computed, it can be visualized by any animated camera. Results present significant high speed-ups showing that the technique could be an interesting alternative to deterministic methods for computing non-interactive radiosity animations for moderately complex scenarios

Calcrete profile development in quaternary alluvial sequences, southeast Spain: Implications for using calcretes as a basis for landform chronologies

A detailed study of the morphology and micro-morphology of Quaternary alluvial calcrete profiles from the Sorbas Basin shows that calcretes may be morphologically simple or complex. The 'simple' profiles reflect pedogenesis occurring after alluvial terrace formation and consist of a single pedogenic horizon near the land surface. The 'complex' profiles reflect the occurrence of multiple calcrete events during terrace sediment aggradation and further periods of pedogenesis after terrace formation. These 'complex' calcrete profiles are consequently described as composite profiles. The exact morphology of the composite profiles depends upon: (1) the number of calcrete-forming events occurring during terrace sediment aggradation; (2) the amount of sediment accretion that occurs between each period of calcrete formation; and (3) the degree of pedogenesis after terrace formation. Simple calcrete profiles are most useful in establishing landform chronologies because they represent a single phase of pedogenesis after terrace formation. Composite profiles are more problematic. Pedogenic calcretes that form within them may inherit carbonate from calcrete horizons occurring lower down in the terrace sediments. In addition erosion may lead to the exhumation of older calcretes within the terrace sediment. Calcrete 'inheritance' may make pedogenic horizons appear more mature than they actually are and produce horizons containing carbonate embracing a range of ages. Calcrete exhumation exposes calcrete horizons whose morphology and radiometric ages are wholly unrelated to terrace surface age. Composite profiles are, therefore, only suitable for chronological studies if the pedogenic horizon capping the terrace sequence can be clearly distinguished from earlier calcrete-forming events. Thus, a detailed morphological/micro-morphological study is required before any chronological study is undertaken. This is the only way to establish whether particular calcrete profiles are suitable for dating purposes. Copyright (C) 2003 John Wiley Sons, Ltd.

Identification of novel expressed sequences, up-regulated in the leucocytes of chronic fatigue syndrome patients

BACKGROUND: Chronic fatigue syndrome (CFS) is an increasing medical phenomenon of unknown aetiology leading to high levels of chronic morbidity. Of the many hypotheses that purport to explain this disease, immune system activation, as a central feature, has remained prominent but unsubstantiated. Supporting this, a number of important cytokines have previously been shown to be over-expressed in disease subjects. The diagnosis of CFS is highly problematic since no biological markers specific to this disease have been identified. The discovery of genes relating to this condition is an important goal in seeking to correctly categorize and understand this complex syndrome. OBJECTIVE: The aim of this study was to screen for changes in gene expression in the lymphocytes of CFS patients. METHODS: 'Differential Display' is a method for comparing mRNA populations for the induction or suppression of genes. In this technique, mRNA populations from control and test subjects can be 'displayed' by gel electrophoresis and screened for differing banding patterns. These differences are indicative of altered gene expression between samples, and the genes that correspond to these bands can be cloned and identified. Differential display has been used to compare expression levels between four control subjects and seven CFS patients. RESULTS: Twelve short expressed sequence tags have been identified that were over-expressed in lymphocytes from CFS patients. Two of these correspond to cathepsin C and MAIL1 - genes known to be upregulated in activated lymphocytes. The expression level of seven of the differentially displayed sequences have been verified by quantifying relative level of these transcripts using TAQman quantitative PCR. CONCLUSION: Taken as a whole, the identification of novel gene tags up-regulated in CFS patients adds weight to the idea that CFS is a disease characterized by subtle changes in the immune system.

Complex history of a chromosomal paralogy region: insights from amphioxus aromatic amino acid hydroxylase genes and insulin-related genes

Aromatic amino acid hydroxylase (AAAH) genes and insulin-like genes form part of an extensive paralogy region shared by human chromosomes 11 and 12, thought to have arisen by tetraploidy in early vertebrate evolution. Cloning of a complementary DNA (cDNA) for an amphioxus (Branchiostoma floridae) hydroxylase gene (AmphiPAH) allowed us to investigate the ancestry of the human chromosome 11/12 paralogy region. Molecular phylogenetic evidence reveals that AmphiPAH is orthologous to vertebrate phenylalanine (PAH) genes; the implication is that all three vertebrate AAAH genes arose early in metazoan evolution, predating vertebrates. In contrast, our phylogenetic analysis of amphioxus and vertebrate insulin-related gene sequences is consistent with duplication of these genes during early chordate ancestry. The conclusion is that two tightly linked gene families on human chromosomes 11 and 12 were not duplicated coincidentally. We rationalize this paradox by invoking gene loss in the AAAH gene family and conclude that paralogous genes shared by paralogous chromosomes need not have identical evolutionary histories.

Characterization of the complex locus of bean encoding polygalacturonase-inhibiting proteins reveals subfunctionalization for defense against fungi and insects.

Polygalacturonase-inhibiting proteins (PGIPs) are extracellular plant inhibitors of fungal endopolygalacturonases (PGs) that belong to the superfamily of Leu-rich repeat proteins. We have characterized the full complement of pgip genes in the bean (Phaseolus vulgaris) genotype BAT93. This comprises four clustered members that span a 50-kb region and, based on their similarity, form two pairs (Pvpgip1/Pvpgip2 and Pvpgip3/Pvpgip4). Characterization of the encoded products revealed both partial redundancy and subfunctionalization against fungal-derived PGs. Notably, the pair PvPGIP3/PvPGIP4 also inhibited PGs of two mirid bugs (Lygus rugulipennis and Adelphocoris lineolatus). Characterization of Pvpgip genes of Pinto bean showed variations limited to single synonymous substitutions or small deletions. A three-amino acid deletion encompassing a residue previously identified as crucial for recognition of PG of Fusarium moniliforme was responsible for the inability of BAT93 PvPGIP2 to inhibit this enzyme. Consistent with the large variations observed in the promoter sequences, reverse transcription-PCR expression analysis revealed that the different family members differentially respond to elicitors, wounding, and salicylic acid. We conclude that both biochemical and regulatory redundancy and subfunctionalization of pgip genes are important for the adaptation of plants to pathogenic fungi and phytophagous insects.

Chaos of Wolbachia Sequences Inside the Compact Fig Syconia of Ficus benjamina (Ficus: Moraceae)

Figs and fig wasps form a peculiar closed community in which the Ficus tree provides a compact syconium (inflorescence) habitat for the lives of a complex assemblage of Chalcidoid insects. These diverse fig wasp species have intimate ecological relationships within the closed world of the fig syconia. Previous surveys of Wolbachia, maternally inherited endosymbiotic bacteria that infect vast numbers of arthropod hosts, showed that fig wasps have some of the highest known incidences of Wolbachia amongst all insects. We ask whether the evolutionary patterns of Wolbachia sequences in this closed syconium community are different from those in the outside world. In the present study, we sampled all 17 fig wasp species living on Ficus benjamina, covering 4 families, 6 subfamilies, and 8 genera of wasps. We made a thorough survey of Wolbachia infection patterns and studied evolutionary patterns in wsp (Wolbachia Surface Protein) sequences. We find evidence for high infection incidences, frequent recombination between Wolbachia strains, and considerable horizontal transfer, suggesting rapid evolution of Wolbachia sequences within the syconium community. Though the fig wasps have relatively limited contact with outside world, Wolbachia may be introduced to the syconium community via horizontal transmission by fig wasps species that have winged males and visit the syconia earlier.

GTP binding controls complex formation by the human ROCO protein MASL1

The human ROCO proteins are a family of multi-domain proteins sharing a conserved ROC-COR supra-domain. The family has four members: leu- cine-rich repeat kinase 1 (LRRK1), leucine-rich repeat kinase 2 (LRRK2), death-associated protein kinase 1 (DAPK1) and malignant fibrous histiocy- toma amplified sequences with leucine-rich tandem repeats 1 (MASL1). Previous studies of LRRK1/2 and DAPK1 have shown that the ROC (Ras of complex proteins) domain can bind and hydrolyse GTP, but the cellular consequences of this activity are still unclear. Here, the first biochemical characterization of MASL1 and the impact of GTP binding on MASL1 complex formation are reported. The results demonstrate that MASL1, similar to other ROCO proteins, can bind guanosine nucleotides via its ROC domain. Furthermore, MASL1 exists in two distinct cellular com- plexes associated with heat shock protein 60, and the formation of a low molecular weight pool of MASL1 is modulated by GTP binding. Finally, loss of GTP enhances MASL1 toxicity in cells. Taken together, these data point to a central role for the ROC/GTPase domain of MASL1 in the reg- ulation of its cellular function.

IntFOLD: an integrated server for modelling protein structures and functions from amino acid sequences

IntFOLD is an independent web server that integrates our leading methods for structure and function prediction. The server provides a simple unified interface that aims to make complex protein modelling data more accessible to life scientists. The server web interface is designed to be intuitive and integrates a complex set of quantitative data, so that 3D modelling results can be viewed on a single page and interpreted by non-expert modellers at a glance. The only required input to the server is an amino acid sequence for the target protein. Here we describe major performance and user interface updates to the server, which comprises an integrated pipeline of methods for: tertiary structure prediction, global and local 3D model quality assessment, disorder prediction, structural domain prediction, function prediction and modelling of protein-ligand interactions. The server has been independently validated during numerous CASP (Critical Assessment of Techniques for Protein Structure Prediction) experiments, as well as being continuously evaluated by the CAMEO (Continuous Automated Model Evaluation) project. The IntFOLD server is available at: http://www.reading.ac.uk/bioinf/IntFOLD/

Resurrection of Two Species From Synonymy of Anopheles (Nyssorhynchus) strodei Root, and Characterization of a Distinct Morphological Form From the Strodei Complex (Diptera: Culicidae)

Anopheles albertoi Unti and Anopheles arthuri Unti are revived from the synonymy with Anopheles strodei Root, and a distinct morphological form (designated in this study as Anopheles CP Form) from the Strodei Complex of Anopheles (Nyssorhynchus) is characterized. The male genitalia of An. arthuri and An. albertoi are described and illustrated for the first time. An. strodei, An. arthuri, and An. albertoi were first distinguished based on scanning electron microphotos of the eggs, and then each egg type was associated with diagnostic characters of the male genitalia. Identification of Anopheles CP Form was based on morphological characters of the male genitalia, characterized and illustrated in this study. Molecular phylogenetic analysis was most clear when an outgroup was not included, in which case using the nuclear white gene, or the white gene in combination with the mitochondrial cytochrome c oxidase subunit I (COI) gene, clearly separated these four taxa. When Anopheles quadrimaculatus Say and Anopheles stephensi Liston were included as an outgroup, combined white and COI data resolved An. strodei and An. albertoi, whereas An. arthuri was not well resolved. The single sequence of Anopheles CP Form was recovered well separated from other groups in all analyses.

Phylogenetic relationships among species of Anopheles (Nyssorhynchus) (Diptera, Culicidae) based on nuclear and mitochondrial gene sequences

Phylogenetic relationships among 21 species of mosquitoes in subgenus Nyssorhynchus were inferred from the nuclear white and mitochondrial NADH dehydrogenase subunit 6 (ND6) genes. Bayestan phylogenetic methods found that none of the three Sections within Nyssorhynchus (Albimanus, Argyritarsis, Myzorhynchella) were supported in all analyses, although Myzorhynchella was found to be monophyletic at the combined genes Within the Albimanus Section the monophyly of the Stroder Subgroup was strongly supported and within the Myzorhynchella Section Anopheles anrunesi and An lutzu formed a strongly supported monophyletic group The epidemiologically significant Albitarsis Complex showed evidence of paraphyly (relative to An lanet-Myzorhynchella) and discordance across gene trees, and the previously synonomized species of An. dunhami and An goeldii were recovered as sister species Finally, there was evidence of complexes in several species, including An antunesi, An deaneorum, and An. strodei (c) 2010 Elsevier B.V. All rights reserved

Curiously composite structures of a retrotransposon and a complex repeat associated with chromosome ends of Rhynchosciara americana (Diptera: Sciaridae)

In Drosophila, telomere retrotransposons counterbalance the loss of telomeric DNA. The exceptional mechanism of telomere recovery characterized in Drosophila has not been found in lower dipterans (Nematocera). However, a retroelement resembling a telomere transposon and termed ""RaTART"" has been described in the nematoceran Rhynchosciara americana. In this work, DNA and protein sequence analyses, DNA cloning, and chromosomal localization of probes obtained either by PCR or by screening a genomic library were carried out in order to examine additional features of this retroelement. The analyses performed raise the possibility that RaTART represents a genomic clone composed of distinct repetitive elements, one of which is likely to be responsible for its apparent enrichment at chromosome ends. RaTART sequence in addition allowed to assess a novel subtelomeric region of R. americana chromosomes that was analyzed in this work after subcloning a DNA fragment from a phage insert. It contains a complex repeat that is located in the vicinity of simple and complex tandem repeats characterized previously. Quantification data suggest that the copy number of the repeat is significantly lower than that observed for the ribosomal DNA in the salivary gland of R. americana. A short insertion of the RaTART was identified in the cloned segment, which hybridized preferentially to subtelomeres. Like RaTART, it displays truncated sequences related to distinct retrotransposons, one of which has a conceptual translation product with significant identity with an endonuclease from a lepidopteran retrotransposon. The composite structure of this DNA stretch probably reflects mobile element activity in the subtelomeric region analyzed in this work.

Complex rearrangements in patients with duplications of MECP2 can occur by fork stalling and template switching

Duplication at the Xq28 band including the MECP2 gene is one of the most common genomic rearrangements identified in neurodevelopmentally delayed males. Such duplications are non-recurrent and can be generated by a non-homologous end joining (NHEJ) mechanism. We investigated the potential mechanisms for MECP2 duplication and examined whether genomic architectural features may play a role in their origin using a custom designed 4-Mb tiling-path oligonucleotide array CGH assay. Each of the 30 patients analyzed showed a unique duplication varying in size from similar to 250 kb to similar to 2.6 Mb. Interestingly, in 77% of these non-recurrent duplications, the distal breakpoints grouped within a 215 kb genomic interval, located 47 kb telomeric to the MECP2 gene. The genomic architecture of this region contains both direct and inverted low-copy repeat (LCR) sequences; this same region undergoes polymorphic structural variation in the general population. Array CGH revealed complex rearrangements in eight patients; in six patients the duplication contained an embedded triplicated segment, and in the other two, stretches of non-duplicated sequences occurred within the duplicated region. Breakpoint junction sequencing was achieved in four duplications and identified an inversion in one patient, demonstrating further complexity. We propose that the presence of LCRs in the vicinity of the MECP2 gene may generate an unstable DNA structure that can induce DNA strand lesions, such as a collapsed fork, and facilitate a Fork Stalling and Template Switching event producing the complex rearrangements involving MECP2.

The Guarguaraz Complex and the Neoproterozoic-Cambrian evolution of southwestern Gondwana: Geochemical signatures and geochronological constraints

The Guarguaraz Complex, in western Argentina, comprises a metasedimentary assemblage, associated with mafic sills and ultramafic bodies intruded by basaltic dikes, which are interpreted as Ordovician dismembered ophiolites. Two kinds of dikes are recognized, a group associated with the metasediments and the other ophiolite-related. Both have N-MORB signatures, with epsilon(Nd) between +3.5 and +8.2, indicating a depleted source, and Grenville model ages between 0.99 and 1.62 Ga. A whole-rock Sm-Nd isochron yielded an age of 655 +/- 76 Ma for these mafic rocks, which is compatible with cianobacteria and acritarchae recognized in the clastic metasedimentary platform sequences, that indicate a Neoproterozoic (Vendian)-Cambrian age of deposition. The Guarguaraz metasedimentary-ophiolitic complex represents, therefore, a remnant of an oceanic basin developed to the west of the Grenville-aged Cuyania terrane during the Neoproterozoic. The southernmost extension of these metasedimentary sequences in Cordon del Portillo might represent part of this platform and not fragments of the Chilenia terrane. An extensional event related to the fragmentation of Rodinia is represented by the mafic and ultramafic rocks. The Devonian docking of Chilenia emplaced remnants of ocean floor and slices of the Cuyania terrane (Las Yaretas Gneisses) in tectonic contact with the Neoproterozoic metasediments, marking the Devonian western border of Gondwana. (C) 2009 Elsevier Ltd. All rights reserved.

Sr-Nd isotopic evidence for crustal contamination in the Niquelandia complex, Goias, Central Brazil

The Niquelandia complex is a Neoproterozoic mafic-ultramafic intrusion resulting from fractional crystallization of primary picritic basalt intrusions. It consists of two layered sequences: a lower and larger one (LS), where four stratigraphic units exhibit an upward decrease of ultramafic layers and increase of gabbroic layers; an upper, smaller sequence (US), separated from LS by a high-temperature shear zone and consisting of two stratigraphic units (gabbros + anorthosites and amphibolites). Nd and Sr isotopic analyses and rare earth element (REE) profiles provide evidence that the complex suffered important crustal contamination. The LS isotopic array trends from a DM region with positive epsilon Nd and moderately positive epsilon Sr towards a field occupied by crustal xenoliths, especially abundant in the upper LS (negative epsilon Nd and large, positive E:Sr). Each LS stratigraphic unit is distinct from the next underlying unit, showing lower epsilon Nd and higher epsilon Sr, suggesting inputs of fresh magma and mixing with the contaminated, residual magma. The US is characterised by a relatively high variation of epsilon Nd and constant epsilon Sr. REE patterns vary within each unit from LREE depleted to LREE enriched in the samples having lower epsilon Nd and higher epsilon Sr. The contamination process has been modelled by using the EC-AFC algorithms from [Spera, F.J., Bohrson, W.A., 2001. Energy-constrained open-system magmatic processes 1: general model and energy-constrained assimilation and fractional crystallization (EC-AFC) formulation. J. Petrology 42, 999-1018]. The differences between the LS and US isotopic arrays are consistent with contamination by the same crustal component, provided that its melting degree was higher in LS than in US. The different degrees of anatexis are explained by the heat budget released from the magma, higher in LS (because of its larger mass) than in US. Comparison of the correlations between isotopes and incompatible trace element ratios of the models and of the gabbros shows some differences, which are demonstrably related with the variable amount of cumulus phases and trapped melt in the gabbros. (c) 2007 Elsevier Ltd. All rights reserved.