Search

The S:A222V point mutation, within the G clade, was characteristic of the 20E (EU1) SARS-CoV-2 variant identified in Spain in early summer 2020. This mutation has now reappeared in the Delta subvariant AY.4.2, raising questions about its specific effect on viral infection. We report combined serological, functional, structural and computational studies characterizing the impact of this mutation. Our results reveal that S:A222V promotes an increased RBD opening and slightly increases ACE2 binding as compared to the parent S:D614G clade. Finally, S:A222V does not reduce sera neutralization capacity, suggesting it does not affect vaccine effectiveness., This research work was supported by the European Commission–NextGenerationEU through the CSIC Global Health Platform. Additionally, authors would like to acknowledge economic support from the Spanish Ministry of Science and Innovation through Grants: PID2019-104757RB-I00 funded by MCIN/AEI/ 10.13039/501100011033, RTI2018-094399-A-I00, and “ERDF A way of making Europe”, by the “European Union”, Grant SEV 2017-0712 funded by MCIN/AEI /10.13039/501100011033, the “Comunidad Autónoma de Madrid" through Grant: S2017/BMD3817, and the European Union (EU) and Horizon 2020 through grants: Marie-Curie Fellowship EnLaCES (MSCA IF 2020, Proposal: 101024130) (to JK), HighResCells (ERC - 2018 - SyG, Proposal: 810057), and iNEXT-Discovery (Proposal: 871037). AM, VR, JB and JLL are funded by CIBERER-ISCIII (proposal: COV20/00437), Fondo Supera COVID-19 (proposal: CSICCOVID19-082), Banco Santander (Proposal: BlockAce), and CSIC PTI Salud Global (Proposal: 202080E110). VR is funded by the Spanish Ministry of Science and Innovation through Grant PID2020-120322RB-C21. IC is funded by project PID2019-104477RB-100, Fondo COVID COV20/00140 and ERC CoG 101001038. MC is funded by the RyC program from the Spanish Ministry of Science and Innovation, the Generalitat Valenciana (SEJI/2019/011).

La mayoría de la información biomédica se encuentra presente como texto libre y existe un aumento constante de la creación de los documentos relacionados al área médica. El problema principal trata de que el texto libre mencionado se encuentra disponible pero en diferentes fuentes, puesto que no está unificado en un único corpus. Si bien es cierto que la lengua inglesa es el idioma preferido para escribir esos documentos, y existen varios corpus consolidando cierta información, terceros idiomas (como por ejemplo, español) que presentan una amplia información médica, aún necesitan desarrollar un corpus. Las tecnologías emergentes de Aprendizaje Automático crean y procesan diferentes métodos constantemente y aunque las técnicas matemáticas complejas son relevantes, los corpus de textos son la base del estudio del lenguaje natural. En ésta tesis de Máster, presentamos un nuevo corpus médico en español denominado COpenMed Corpus. El listado original de las páginas web (URLs) empleadas para construir el corpus alcanza la cantidad de 19.571 recursos relacionados con un total de 9.099 entidades. Una función especial de relleno de recursos relacionados a una misma identidad también se ha añadido: la función de relleno se emplea para establecer un valor N referente al mínimo número de referencias por entidad. Para la construcción del corpus se ha empleado N = 10 alcanzando un total de 90.990 páginas web descargadas. Para ésta descarga, se ha desarrollado un método que hace uso de técnicas Webscraping para descargar y procesar las páginas web obtenidas del listado. Como preparativos para un posible futuro trabajo y futuros pasos de uso del corpus creado, se ha pre-entrando dos modelos diferentes de Procesamiento Natural del Lenguaje. Los modelos empleados han sido RoBERTa y XLNet, ambos pre-entrenados con COpenMed corpus.---ABSTRACT---The majority of biomedical information is present as free text and there is a constant augment of documents created in a daily basis for this medical field. The main problem is that all of the free text is available but located at different sources, since it is not unified in a single corpus. Although English is the preferred language for writing documents, and has several corpus consolidating the information, another languages with vast amount of information has yet to develop a large corpus. Emerging Machine Learning technologies are appealing and different methods and processes are tested in a constant basis. Even though these complex mathematical techniques are relevant, text corpora are the basis of natural language studying. In this Master thesis we present a new large Spanish medical corpus: COpenMed Corpus. Original pool of URLs employed for constructing the corpus reaches the amount of 19.571 resources related to a total of 9.099 entities. A padding method for resources referred to a same entity was also added, the padding is employed by establishing a N value for minimum number of resources per entity. We employed N = 10 reaching a total of 90.990 URLs scraped. For this goal, we develop a method that employs Webscraping techniques for downloading and processing a substantial amount of URLs from a given pool of URLs. As preparations for further steps and future work, pre-training of two different Natural Language Processing models was also done in this Master thesis. Models employed are RoBERTa and XLNet, both of them pre-trained with COpenMed corpus.

The development of the electron microscopy (EM) has led to new algorithms devoted to the analysis of the images obtained by the EM know as micrographs. These images are characterized for being noisy, making the extraction of information a difficult task. To that end, several reconstruction packages are being developed that provide with powerful tools to denoise, extract and analyze the micrographs to obtain the 3D model (volume) of the sample. Since there is no a gold standard that can be followed to process all micrographs in a single way, researches rely on the combination of algorithms of several packages to get the optimal result. However, the lack of consensus between the packages makes difficult to combine different programs. Also, the algorithms should be made as simple and automatic as possible to facilitate the task of the researchers without degrading the quality of the results. The purpose of this bachelor thesis is to introduce required algorithms that can be used to reconstruct the 3D model of a sample out of the micrographs acquired by the electron microscope. The analysis will be focused on two individual and complementary tasks: the development of a mathematical basis aimed at the automatization and simplification of the deformation calculations carried out to optimize the 3D templates used to reconstruct new 3D models, and the protocolization of a package that uses ab initio methods to avoid the requirements of a 3D template to reconstruct a model. The algorithms will be tested in order to validate and analyze their performance, so they can be used in real problems and applications. Ingeniería Biomédica

Tesis doctoral inédita leída en la Universidad Autónoma de Madrid, Escuela Politécnica Superior, Departamento de Ingeniería Informática. Fecha de lectura: 22-06-2016, Structural information of macromolecular complexes provides key insights into the way they carry out their biological functions. In turn, Electron microscopy (EM) is an essential tool to study the structure and function of biological macromolecules at a medium-high resolution. In this context, Single-Particle Analysis (SPA), as an EM modality, is able to yield Three-Dimensional (3-D) structural information for large biological complexes at near atomic resolution by combining many thousands of projection images. However, these views su er from low Signal-to-Noise Ratios (SNRs), since an extremely low total electron dose is used during exposure to reduce radiation damage and preserve the functional structure of macromolecules. In recent years, the emergence of Direct Detection Devices (DDDs) has opened up the possibility of obtaining images with higher SNRs. These detectors provide a set of frames instead of just one micrograph, which makes it possible to study the behavior of frozen hydrated specimens as a function of electron dose and rate. In this way, it has become apparent that biological specimens embedded in a solid matrix of amorphous ice move during imaging, resulting in Beam-Induced Motion (BIM). Therefore, alignment of frames should be added to the classical standard data processing work ow of single-particle reconstruction, which includes: particle selection, particle alignment, particle classi cation, 3-D reconstruction, and model re nement. In this thesis, we propose new algorithms and improvements for three important steps of this work ow: movie alignment, particles selection, and 3-D reconstruction. For movie alignment, a methodology based on a robust to noise optical ow approach is proposed that can e ciently correct for local movements and provide quantitative analysis of the BIM pattern. We then introduce a method for automatic particle selection in micrographs that uses some new image features to train two classi ers to learn from the user the kind of particles he is interested in. Finally, for 3-D reconstruction, we introduce a gridding-based direct Fourier method that uses a weighting technique to compute a uniform sampled Fourier transform. The algorithms are fully implemented in the open-source Xmipp package (http://xmipp.cnb.csic.es)