No description available.

During the last decades, mitochondria and their DNA have become a hot topic of research due to their essential roles which are necessary for cells survival and pathology. In this study, multiple methods have been developed to help with the understanding of mitochondrial DNA and its evolution. These methods tackle two essential problems in this area: the accurate annotation of protein-coding genes and mitochondrial genome rearrangements. Mitochondrial genome sequences are published nowadays with increasing pace, which creates the need for accurate and fast annotation tools that do not require manual intervention. In this work, an automated pipeline for fast de-novo annotation of mitochondrial protein-coding genes is implemented. The pipeline includes methods for enhancing multiple sequence alignment, detecting frameshifts and building protein profiles guided by phylogeny. The methods are tested on animal mitogenomes available in RefSeq, the comparison with reference annotations highlights the high quality of the produced annotations. Furthermore, the frameshift method predicted a large number of frameshifts, many of which were unknown. Additionally, an efficient partially-local alignment method to investigate genomic rearrangements in mitochondrial genomes is presented in this study. The method is novel and introduces a partially-local dynamic programming algorithm on three sequences around the breakpoint region. Unlike the existing methods which study the rearrangement at the genes order level, this method allows to investigate the rearrangement on the molecular level with nucleotides precision. The algorithm is tested on both artificial data and real mitochondrial genomic sequences. Surprisingly, a large fraction of rearrangements involve the duplication of local sequences. Since the implemented approach only requires relatively short parts of genomic sequence around a breakpoint, it should be applicable to non-mitochondrial studies as well.

The Internet provides a powerful medium for communication and collaboration. The ability one has to connect and interact with web-based tools from anywhere in the world makes the Internet ideal for such tasks. However, the lack of native tools can be a hindrance when deploying collaborative initiatives, as many current projects require specialized software in order to operate. This thesis demonstrates, with the comparably recent advances in browser technology and Document Object Model (DOM) implementation, a web-based collaborative annotation system can be developed that can be accessed by a user through a standards-compliant web browser. Such a system, demonstrated to work on the commonly-used web browsers constituting the vast majority of web traffic, was implemented using open-source tools and industry-recognized standards. Additionally, it accepts static copies of most standard document formats for both handwritten and typed annotations, while maintaining an archived copy of the original. The system developed for this thesis lends itself to use in a number of different process domains, as most collaborative annotation approaches can be described by a single process model. While a number of possible usage scenarios are discussed, this thesis approaches system usage only in an academic setting, focusing on applicability of the system to electronic grading and document exchange. From here, additional system usage can be easily extrapolated.

By observing human driving with a “digital head" (combined video camera and accelerometers) and taking a few hand annotations, we can automatically annotate regions in a robot's field of view that should be interpreted as obstacles to be avoided. This is accomplished by detecting the movement for a given frame in a video. Some hand annotations of video frames are necessary and they are used to create Probability Grids. Using the movement data and the Probability Grids, it is possible to annotate large amounts of video data quickly in an automated system.

Single-cell RNA-sequencing (scRNA-seq) technology enables researchers to investigate a genome at the cellular level with unprecedented resolution. An organism consists of a heterogeneous collection of cell types, each of which plays a distinct role in various biological processes. Hence, the first step of scRNA-seq data analysis often is to distinguish cell types so that they can be investigated separately. Researchers have recently developed several automated cell type annotation tools based on supervised machine learning algorithms, requiring neither biological knowledge nor subjective human decisions. Dropout is a crucial characteristic of scRNA-seq data which is widely utilized in differential expression analysis but not by existing cell annotation methods. We present scAnnotate, a cell annotation tool that fully utilizes dropout information. We model every gene’s marginal distribution using a mixture model, which describes both the dropout proportion and the distribution of the non-dropout expression levels. Then, using an ensemble machine learning approach, we combine the mixture models of all genes into a single model for cell-type annotation. This combining approach can avoid estimating numerous parameters in the high-dimensional joint distribution of all genes. Using fourteen real scRNA-seq datasets, we demonstrate that scAnnotate is competitive against nine existing annotation methods, and that it accurately annotates cells when training and test data are (1) similar, (2) cross-platform, and (3) cross-species. Of the cells that are incorrectly annotated by scAnnotate, we find that a majority are different from those of other methods. / Graduate / 2023-07-27

With the proliferation of social media, gathering data has became cheaper and easier than before. However, this data can not be used for supervised machine learning without labels. Asking experts to annotate sufficient data for training is both expensive and time-consuming. Current techniques provide two solutions to reducing the cost and providing sufficient labels: crowdsourcing and active learning. Crowdsourcing, which outsources tasks to a distributed group of people, can be used to provide a large quantity of labels but controlling the quality of labels is hard. Active learning, which requires experts to annotate a subset of the most informative or uncertain data, is very sensitive to the annotation errors. Though these two techniques can be used independently of one another, by using them in combination they can complement each other’s weakness. In this thesis, I investigate the development of active learning Support Vector Machines (SVMs) and expand this model to sequential data. Then I discuss the weakness of combining active learning and crowdsourcing, since the active learning is very sensitive to low quality annotations which are unavoidable for labels collected from crowdsourcing. In this thesis, I propose three possible strategies, incremental relabeling, importance-weighted label prediction and active Bayesian Networks. The incremental relabeling strategy requires workers to devote more annotations to uncertain samples, compared to majority voting which allocates different samples the same number of labels. Importance-weighted label prediction employs an ensemble of classifiers to guide the label requests from a pool of unlabeled training data. An active learning version of Bayesian Networks is used to model the difficulty of samples and the expertise of workers simultaneously to evaluate the relative weight of workers’ labels during the active learning process. All three strategies apply different techniques with the same expectation – identifying the optimal solution for applying an active learning model with mixed label quality to iii crowdsourced data. However, the active Bayesian Networks model, which is the core element of this thesis, provides additional benefits by estimating the expertise of workers during the training phase. As an example application, I also demonstrate the utility of crowdsourcing for human activity recognition problems.

No description available.

No description available.

No description available.

La technologie Web 2.0 a donné lieu à un large éventail de plates-formes de partage de photos. Il est désormais possible d’annoter des images de manière collaborative, au moyen de mots-clés; ce qui permet une gestion et une recherche efficace de ces images. Toutefois, l’annotation manuelle est laborieuse et chronophage. Au cours des dernières années, le nombre grandissant de photos annotées accessibles sur le Web a permis d'expérimenter de nouvelles méthodes d'annotation automatique d'images. L'idée est d’identifier, dans le cas d’une photo non annotée, un ensemble d'images visuellement similaires et, a fortiori, leurs mots-clés, fournis par la communauté. Il existe actuellement un nombre considérable de photos associées à des informations de localisation, c'est-à-dire géo-localisées. Nous exploiterons, dans le cadre de cette thèse, ces informations et proposerons une nouvelle approche pour l'annotation automatique d'images géo-localisées. Notre objectif est de répondre aux principales limites des approches de l'état de l'art, particulièrement concernant la qualité des annotations produites ainsi que la rapidité du processus d'annotation. Tout d'abord, nous présenterons une méthode de collecte de données annotées à partir du Web, en se basant sur la localisation des photos et les liens sociaux entre leurs auteurs. Par la suite, nous proposerons une nouvelle approche afin de résoudre l’ambiguïté propre aux tags d’utilisateurs, le tout afin d’assurer la qualité des annotations. L'approche démontre l'efficacité de l'algorithme de recherche de caractéristiques discriminantes, dit de Laplace, dans le but d’améliorer la représentation de l'annotation. En outre, une nouvelle mesure de distance entre mots-clés sera présentée, qui étend la divergence de Jensen-Shannon en tenant compte des fluctuations statistiques. Dans le but d'identifier efficacement les images visuellement proches, la thèse étend sur deux point l'algorithme d'état de l'art en comparaison d'images, appelé SURF (Speeded-Up Robust Features). Premièrement, nous présenterons une solution pour filtrer les points-clés SURF les plus significatifs, au moyen de techniques de classification, ce qui accélère l'exécution de l'algorithme. Deuxièmement, la précision du SURF sera améliorée, grâce à une comparaison itérative des images. Nous proposerons une un modèle statistique pour classer les annotations récupérées selon leur pertinence du point de vue de l'image-cible. Ce modèle combine différents critères, il est centré sur la règle de Bayes. Enfin, l'efficacité de l'approche d'annotation ainsi que celle des contributions individuelles sera démontrée expérimentalement. / In the Web 2.0 era, platforms for sharing and collaboratively annotating images with keywords, called tags, became very popular. Tags are a powerful means for organizing and retrieving photos. However, manual tagging is time consuming. Recently, the sheer amount of user-tagged photos available on the Web encouraged researchers to explore new techniques for automatic image annotation. The idea is to annotate an unlabeled image by propagating the labels of community photos that are visually similar to it. Most recently, an ever increasing amount of community photos is also associated with location information, i.e., geotagged. In this thesis, we aim at exploiting the location context and propose an approach for automatically annotating geotagged photos. Our objective is to address the main limitations of state-of-the-art approaches in terms of the quality of the produced tags and the speed of the complete annotation process. To achieve these goals, we, first, deal with the problem of collecting images with the associated metadata from online repositories. Accordingly, we introduce a strategy for data crawling that takes advantage of location information and the social relationships among the contributors of the photos. To improve the quality of the collected user-tags, we present a method for resolving their ambiguity based on tag relatedness information. In this respect, we propose an approach for representing tags as probability distributions based on the algorithm of Laplacian Score feature selection. Furthermore, we propose a new metric for calculating the distance between tag probability distributions by extending Jensen-Shannon Divergence to account for statistical fluctuations. To efficiently identify the visual neighbors, the thesis introduces two extensions to the state-of-the-art image matching algorithm, known as Speeded Up Robust Features (SURF). To speed up the matching, we present a solution for reducing the number of compared SURF descriptors based on classification techniques, while the accuracy of SURF is improved through an efficient method for iterative image matching. Furthermore, we propose a statistical model for ranking the mined annotations according to their relevance to the target image. This is achieved by combining multi-modal information in a statistical framework based on Bayes' Rule. Finally, the effectiveness of each of mentioned contributions as well as the complete automatic annotation process are evaluated experimentally.