<div>Deep neural networks (DNNs) have gained tremendous attention over the recent years due to their outstanding performance in solving many problems in different fields of science and technology. Currently, this field is of interest to many researchers and growing rapidly. The ability of DNNs to learn new concepts with minimal instructions facilitates applying current DNN-based methods to new problems. Here in this dissertation, three methods based on DNNs are discussed, tackling different problems in the field of computer-aided drug design.</div><div><br></div><div>The first method described addresses the problem of prediction of hydration properties from 3D structures of proteins without requiring molecular dynamics simulations. Water plays a major role in protein-ligand interactions and identifying (de)solvation contributions of water molecules can assist drug design. Two different model architectures are presented for the prediction the hydration information of proteins. The performance of the methods are compared with other conventional methods and experimental data. In addition, their applications in ligand optimization and pose prediction is shown.</div><div><br></div><div>The design of de novo molecules has always been of interest in the field of drug discovery. The second method describes a generative model that learns to derive features from protein sequences to design de novo compounds. We show how the model can be used to generate molecules similar to the known for the targets the model have not seen before and compare with benchmark generative models.</div><div><br></div><div>Finally, it is demonstrated how DNNs can learn to predict secondary structure propensity values derived from NMR ensembles. Secondary structure propensities are important in identifying flexible regions in proteins. Protein flexibility has a major role in drug-protein binding, and identifying such regions can assist in development of methods for ligand binding prediction. The prediction performance of the method is shown for several proteins with two or more known secondary structure conformations.</div>

Die vorliegende Dissertation befasst sich mit der theoretischen Vorhersage neuer Materialien. Ein evolutionärer Algorithmus, der zur Lösung dieses globalen Optimierungsproblems Konzepte der natürlichen Evolution imitiert, wurde entwickelt und ist als Programmpaket EVO frei verfügbar. EVO findet zuverlässig sowohl bekannte als auch neuartige Kristallstrukturen. Beispielsweise wurden die Strukturen von Germaniumnitrofluorid, einer neue Borschicht und mit dem gekreuzten Graphen einer bisher unbekannte Kohlenstoffstruktur gefunden. Ferner wurde in der Arbeit gezeigt, dass das reine Auffinden solcher Strukturen der erste Teil einer erfolgreichen Vorhersage ist. Weitere aufwendige Berechnungen sind nötig, die Aufschluss über die Stabilität der hypothetischen Struktur geben und Aussagen über zu erwartende Materialeigenschaften liefern.

No description available.

Proteine spielen auf der zellulären Ebene eines Organismus eine fundamentale Rolle. Sie sind quasi die „Maschinen“ der Zelle. Ihre Bedeutung wird nicht zuletzt in ihrem Namen deutlich, welcher 1838 erstmals von J. Berzelius verwendet wurde und „das Erste“, „das Wichtigste“ bedeutet. Proteine sind aus Aminosäuren aufgebaute Moleküle. Unter physiologischen Bedingungen besitzen sie eine definierte dreidimensionale Gestalt, welche für ihre biologische Funktion bestimmend ist. Es wird heutzutage davon ausgegangen, dass diese dreidimensionale, stabile Struktur von Proteinen eindeutig durch die Abfolge der einzelnen Aminosäuren, der Sequenz, bestimmt ist. Diese Abfolge ist für jedes Protein in der Desoxyribonukleinsäure (DNS) gespeichert. Es ist allerdings eines der größten ungelösten Probleme der letzten Jahrzehnte, wie die Beziehung zwischen Sequenz und 3D-Struktur tatsächlich aussieht. Die Beantwortung dieser Fragestellung erfordert interdisziplinäre Ansätze aus Biologie, Informatik und Physik. In dieser Arbeit werden mit Hilfe von Methoden der theoretischen (Bio-) Physik einige der damit verbundenen Aspekte untersucht. Das Hauptaugenmerk liegt dabei auf Wechselwirkungen der einzelnen Aminosäuren eines Proteins untereinander, wofür in dieser Arbeit ein entsprechendes Energiemodell entwickelt wurde. Es werden Grundzustände sowie Energielandschaften untersucht und mit experimentellen Daten verglichen. Die Stärke der Wechselwirkung einzelner Aminosäuren erlaubt zusätzlich Aussagen über die Stabilität von Proteinen bezüglich mechanischer Kräfte. Die vorliegende Arbeit unterteilt sich wie folgt: Kapitel 2 dient der Einleitung und stellt Proteine und ihre Funktionen dar. Kapitel 3 stellt die Modellierung der Proteinstrukturen in zwei verschiedenen Modellen vor, welche in dieser Arbeit entwickelt wurden, um 3D-Strukturen von Proteinen zu beschreiben. Anschließend wird in Kapitel 4 ein Algorithmus zum Auffinden des exakten Energieminimums dargestellt. Kapitel 5 beschäftigt sich mit der Frage, wie eine geeignete diskrete Energiefunktion aus experimentellen Daten gewonnen werden kann. In Kapitel 6 werden erste Ergebnisse dieses Modells dargestellt. Der Frage, ob der experimentell bestimmte Zustand dem energetischen Grundzustand eines Proteins entspricht, wird in Kapitel 7 nachgegangen. Die beiden Kapitel 8 und 9 zeigen die Anwendung des Modells an zwei Proteinen, dem Tryptophan cage protein als dem kleinsten, stabilen Protein und Kinesin, einem Motorprotein, für welches 2007 aufschlussreiche Experimente zur mechanischen Stabilität durchgeführt wurden. Kapitel 10 bis 12 widmen sich Membranproteinen. Dabei beschäftigt sich Kapitel 10 mit der Vorhersage von stabilen Bereichen (sog. Entfaltungsbarrieren) unter externer Krafteinwirkung. Zu Beginn wird eine kurze Einleitung zu Membranproteinen gegeben. Im folgenden Kapitel 11 wird die Entfaltung mit Hilfe des Modells und Monte-Carlo-Techniken simuliert. Mit dem an Membranproteine angepassten Wechselwirkungsmodell ist es möglich, den Einfluss von Mutationen auch ohne explizite strukturelle Informationen vorherzusagen. Dieses Thema wird in Kapitel 12 diskutiert. Die Beziehung zwischen Primär- und Tertiärstruktur eines Proteins wird in Kapitel 13 behandelt. Es wird ein Ansatz skizziert, welcher in der Lage ist, Strukturbeziehungen zwischen Proteinen zu detektieren, die mit herkömmlichen Methoden der Bioinformatik nicht gefunden werden können. Die letzten beiden Kapitel schließlich geben eine Zusammenfassung bzw. einen Ausblick auf künftige Entwicklungen und Anwendungen des Modells.

No description available.

Accurate predictions of binding free energies from computer simulations are an invaluable resource for understanding biochemical processes and drug action. The primary aim of the work described in the thesis was to predict and understand ligand binding to several proteins of major pharmaceutical importance using computational methods. We report a computational strategy to quantitatively predict the effects of alanine scanning and ligand modifications based on molecular dynamics free energy simulations. A smooth stepwise scheme for free energy perturbation calculations is derived and applied to a series of thirteen alanine mutations of the human neuropeptide Y1 G-protein coupled receptor and a series of eight analogous antagonists. The robustness and accuracy of the method enables univocal interpretation of existing mutagenesis and binding data. We show how these calculations can be used to validate structural models and demonstrate their ability to discriminate against suboptimal ones. Site-directed mutagenesis, homology modelling and docking were further used to characterize agonist binding to the human neuropeptide Y2 receptor, which is important in feeding behavior and an obesity drug target.  In a separate project, homology modelling was also used for rationalization of mutagenesis data for an integron integrase involved in antibiotic resistance. Blockade of the hERG potassium channel by various drug-like compounds, potentially causing serious cardiac side effects, is a major problem in drug development. We have used a homology model of hERG to conduct molecular docking experiments with a series of channel blockers, followed by molecular dynamics simulations of the complexes and evaluation of binding free energies with the linear interaction energy method. The calculations are in good agreement with experimental binding affinities and allow for a rationalization of three-dimensional structure-activity relationships with implications for design of new compounds. Docking, scoring, molecular dynamics, and the linear interaction energy method were also used to predict binding modes and affinities for a large set of inhibitors to HIV-1 reverse transcriptase. Good agreement with experiment was found and the work provides a validation of the methodology as a powerful tool in structure-based drug design. It is also easily scalable for higher throughput of compounds.

Membrane proteins represent about one third of all known vertebrate proteins and over half of the current drug targets. Knowledge of their three-dimensional (3D) structure is worth millions of pounds to the pharmaceutical industry. Yet experimental structure elucidation of membrane proteins is a slow and expensive process. In the absence of experimental data, computational modelling tools can be used to close the gap between the numbers of known protein sequences and structures. However, currently available structure prediction tools were developed with globular soluble proteins in mind and perform poorly on membrane proteins. This thesis describes the development of a modelling approach able to predict accurately the structure of transmembrane domains of proteins. In this thesis we build a template-based modelling framework especially for membrane proteins, which uses membrane protein-specific information to inform the modelling process.Firstly, we develop a tool to accurately determine a given membrane protein structure's orientation within the membrane. We offer an analysis of the preferred substitution patterns within the membrane, as opposed to non-membrane environments, and how these differences influence the structures observed. This information is then used to build a set of tools that produce better sequence alignments of membrane proteins, compared to previously available methods, as well as more accurate predictions of their 3D structures. Each chapter describes one new piece of software or information and uses the tools and knowledge described in previous chapters to build up to a complete accurate model of a transmembrane domain.

Contemporary approaches to natural language processing are predominantly based on statistical machine learning from large amounts of text, which has been manually annotated with the linguistic structure of interest. However, such complete supervision is currently only available for the world's major languages, in a limited number of domains and for a limited range of tasks. As an alternative, this dissertation considers methods for linguistic structure prediction that can make use of incomplete and cross-lingual supervision, with the prospect of making linguistic processing tools more widely available at a lower cost. An overarching theme of this work is the use of structured discriminative latent variable models for learning with indirect and ambiguous supervision; as instantiated, these models admit rich model features while retaining efficient learning and inference properties. The first contribution to this end is a latent-variable model for fine-grained sentiment analysis with coarse-grained indirect supervision. The second is a model for cross-lingual word-cluster induction and the application thereof to cross-lingual model transfer. The third is a method for adapting multi-source discriminative cross-lingual transfer models to target languages, by means of typologically informed selective parameter sharing. The fourth is an ambiguity-aware self- and ensemble-training algorithm, which is applied to target language adaptation and relexicalization of delexicalized cross-lingual transfer parsers. The fifth is a set of sequence-labeling models that combine constraints at the level of tokens and types, and an instantiation of these models for part-of-speech tagging with incomplete cross-lingual and crowdsourced supervision. In addition to these contributions, comprehensive overviews are provided of structured prediction with no or incomplete supervision, as well as of learning in the multilingual and cross-lingual settings. Through careful empirical evaluation, it is established that the proposed methods can be used to create substantially more accurate tools for linguistic processing, compared to both unsupervised methods and to recently proposed cross-lingual methods. The empirical support for this claim is particularly strong in the latter case; our models for syntactic dependency parsing and part-of-speech tagging achieve the hitherto best published results for a wide number of target languages, in the setting where no annotated training data is available in the target language.

O processo de auto-organização da estrutura proteica a partir da cadeia de aminoácidos é conhecido como enovelamento. Apesar de conhecermos a estrutura tridimencional de muitas proteínas, para a maioria delas, não possuímos uma compreensão suficiente para descrever em detalhes como a estrutura se organiza a partir da sequência de aminoácidos. É bem conhecido que a formação de núcleos de estruturas locais, conhecida como estrutura secundária, apresenta papel fundamental no enovelamento final da proteína. Desta forma, o desenvolvimento de métodos que permitam não somente predizer a estrutura secundária adotada por um dado resíduo, mas também, a maneira como esse processo deve ocorrer ao longo do tempo é muito relevante em várias áreas da biologia estrutural. Neste trabalho, desenvolvemos dois métodos de predição de estruturas secundárias utilizando modelos com o potencial de fornecer informações mais detalhadas sobre o processo de predição. Um desses modelos foi construído utilizando autômatos celulares, um tipo de modelo dinâmico onde é possível obtermos informações espaciais e temporais. O outro modelo foi desenvolvido utilizando redes neurais residuais profundas. Com este modelo é possível extrair informações espaciais e probabilísticas de suas múltiplas camadas internas de convolução, o que parece refletir, em algum sentido, os estados de formação da estrutura secundária durante o enovelamento. A acurácia da predição obtida por esse modelo foi de ~78% para os resíduos que apresentaram consenso na estrutura atribuída pelos métodos DSSP, STRIDE, KAKSI e PROSS. Tal acurácia, apesar de inferior à obtida pelo PSIPRED, o qual utiliza matrizes PSSM como entrada, é superior à obtida por outros métodos que realizam a predição de estruturas secundárias diretamente a partir da sequência de aminoácidos. / The process of self-organization of the protein structure is known as folding. Although we know the structure of many proteins, for a majority of them, we do not have enough understanding to describe in details how the structure is organized from its amino acid sequence. In this work, we developed two methods for secondary structure prediction using models that have the potential to provide detailed information about the prediction process. One of these models was constructed using cellular automata, a type of dynamic model where it is possible to obtain spatial and temporal information. The other model was developed using deep residual neural networks. With this model it is possible to extract spatial and probabilistic information from its multiple internal layers of convolution. The accuracy of the prediction obtained by this model was ~ 78% for residues that showed consensus in the structure assigned by the DSSP, STRIDE, KAKSI and PROSS methods. Such value is higher than that obtained by other methods which perform the prediction of secondary structures from the amino acid sequence only.

Submitted by Marlene Santos (marlene.bc.ufg@gmail.com) on 2016-08-05T17:38:42Z No. of bitstreams: 2 Dissertação - Alexandre Barbosa de Almeida - 2016.pdf: 11943401 bytes, checksum: 94f2e941bbde05e098c40f40f0f2f69c (MD5) license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) / Approved for entry into archive by Luciana Ferreira (lucgeral@gmail.com) on 2016-08-09T11:57:53Z (GMT) No. of bitstreams: 2 Dissertação - Alexandre Barbosa de Almeida - 2016.pdf: 11943401 bytes, checksum: 94f2e941bbde05e098c40f40f0f2f69c (MD5) license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) / Made available in DSpace on 2016-08-09T11:57:53Z (GMT). No. of bitstreams: 2 Dissertação - Alexandre Barbosa de Almeida - 2016.pdf: 11943401 bytes, checksum: 94f2e941bbde05e098c40f40f0f2f69c (MD5) license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) Previous issue date: 2016-06-17 / Conselho Nacional de Pesquisa e Desenvolvimento Científico e Tecnológico - CNPq / Proteins are vital for the biological functions of all living beings on Earth. However, they only have an active biological function in their native structure, which is a state of minimum energy. Therefore, protein functionality depends almost exclusively on the size and shape of its native conformation. However, less than 1% of all known proteins in the world has its structure solved. In this way, various methods for determining protein structures have been proposed, either in vitro or in silico experiments. This work proposes a new in silico method called Monte Carlo with Dominance, which addresses the problem of protein structure prediction from the point of view of ab initio and multi-objective optimization, considering both protein energetic and structural aspects. The software GROMACS was used for the ab initio treatment to perform Molecular Dynamics simulations, while the framework ProtPred-GROMACS (2PG) was used for the multi-objective optimization problem, employing genetic algorithms techniques as heuristic solutions. Monte Carlo with Dominance, in this sense, is like a variant of the traditional Monte Carlo Metropolis method. The aim is to check if protein tertiary structure prediction is improved when structural aspects are taken into account. The energy criterion of Metropolis and energy and structural criteria of Dominance were compared using RMSD calculation between the predicted and native structures. It was found that Monte Carlo with Dominance obtained better solutions for two of three proteins analyzed, reaching a difference about 53% in relation to the prediction by Metropolis. / As proteínas são vitais para as funções biológicas de todos os seres na Terra. Entretanto, somente apresentam função biológica ativa quando encontram-se em sua estrutura nativa, que é o seu estado de mínima energia. Portanto, a funcionalidade de uma proteína depende, quase que exclusivamente, do tamanho e da forma de sua conformação nativa. Porém, de todas as proteínas conhecidas no mundo, menos de 1% tem a sua estrutura resolvida. Deste modo, vários métodos de determinação de estruturas de proteínas têm sido propostos, tanto para experimentos in vitro quanto in silico. Este trabalho propõe um novo método in silico denominado Monte Carlo com Dominância, o qual aborda o problema da predição de estrutura de proteínas sob o ponto de vista ab initio e de otimização multiobjetivo, considerando, simultaneamente, os aspectos energéticos e estruturais da proteína. Para o tratamento ab initio utiliza-se o software GROMACS para executar as simulações de Dinâmica Molecular, enquanto que para o problema da otimização multiobjetivo emprega-se o framework ProtPred-GROMACS (2PG), o qual utiliza algoritmos genéticos como técnica de soluções heurísticas. O Monte Carlo com Dominância, nesse sentido, é como uma variante do tradicional método de Monte Carlo Metropolis. Assim, o objetivo é o de verificar se a predição da estrutura terciária de proteínas é aprimorada levando-se em conta também os aspectos estruturais. O critério energético de Metropolis e os critérios energéticos e estruturais da Dominância foram comparados empregando o cálculo de RMSD entre as estruturas preditas e as nativas. Foi verificado que o método de Monte Carlo com Dominância obteve melhores soluções para duas de três proteínas analisadas, chegando a cerca de 53% de diferença da predição por Metropolis.