Computer graphics has become an integral part of various industries such as entertainment (i.e.,films and content creation), fashion (i.e.,virtual try-on), and video games. Computer graphics has evolved tremendously over the past years. It has shown remarkable image generation improvement from low-quality, pixelated images with limited details to highly realistic images with fine details that can often be mistaken for real images. However, the traditional pipeline of rendering an image in computer graphics is complex and time- consuming. The whole process of creating the geometry, material, and textures requires not only time but also significant expertise. In this work, we aim to replace this complex traditional computer graphics pipeline with a simple machine learning model. This machine learning model can synthesize realistic images without requiring expertise or significant time and effort. Specifically, we address the problem of controllable image synthesis. We propose several approaches that allow the user to synthesize realistic content and manipulate images to achieve their desired goals with ease and flexibility. / Doctor of Philosophy / Computer graphics has become an integral part of various industries such as entertainment (i.e.,films and content creation), fashion (i.e.,virtual try-on), and video games. Computer graphics has evolved tremendously over the past years. It has shown remarkable image generation improvement from low-quality, pixelated images with limited details to highly realistic images with fine details that can often be mistaken for real images. However, the traditional process of generating an image in computer graphics is complex and time- consuming. You need to set up a camera and light, and create objects with all sorts of details. This requires not only time but also significant expertise. In this work, we aim to replace this complex traditional computer graphics pipeline with a simple machine learning model. This machine learning model can generate realistic images without requiring expertise or significant time and effort. Specifically, we address the problem of controllable image synthesis. We propose several approaches that allow the user to synthesize realistic content and manipulate images to achieve their desired goals with ease and flexibility.

京都大学 / 新制・課程博士 / 博士(農学) / 甲第24673号 / 農博第2556号 / 新制||農||1099(附属図書館) / 学位論文||R5||N5454(農学部図書室) / 京都大学大学院農学研究科応用生命科学専攻 / (主査)教授 矢﨑 一史, 教授 梅澤 俊明, 教授 伊福 健太郎 / 学位規則第4条第1項該当 / Doctor of Agricultural Science / Kyoto University / DGAM

Advances in genome editing, human induced pluripotent stem cells (iPSC), and cardiac tissue engineering have significantly improved the ability of in vitro models to model cardiac disease. The objective of this dissertation is to leverage cardiac tissue engineering to generate meaningful biological insights into human genetic cardiomyopathies. First, we studied a novel, de novo mutation in the filamin C (FLNC) gene which causes restrictive cardiomyopathy in a young patient. Using engineered cardiac tissues, we showed that this mutation causes a restrictive phenotype marked by increased passive tension and slowed contraction velocities. Complementing our engineered tissues, we used high-throughput calcium imaging to identify compounds which improved myocardial relaxation in mutant cardiomyocytes. These compounds improved function of mutant cardiac tissues, suggesting a potentially targetable pathway in the patient’s mutation. In another study, engineered cardiac tissues and stem cells were used to study BAG3, a dilated cardiomyopathy- related gene, in cardiac fibroblasts. BAG3-/- and wild-type iPSCs were differentiated to cardiac fibroblasts and cardiomyocytes. By generating fully isogenic cardiac tissues and altering cellular genotypes, we determined that the loss of BAG3 in cardiac fibroblasts was deleterious to cardiac tissue function despite genetically normal cardiomyocytes. Further work studying cardiac fibroblasts revealed a mechanistic function of BAG3 in regulating cardiac fibroblast extracellular matrix synthesis. Together, this work highlights the ability of cardiac tissues and stem cells to unravel the complexities of genetic heart disease.

Although the understanding of genetic predisposition to prostate cancer (PCa) has been improved through genome-wide association studies (GWAS), little is known about the biological implication of germline variants residing in coding or non-coding regions in cancer development and progression. Our hypothesis is that inherited variants may predispose to specific early recurrent genomic events observed in PCa adenocarcinomas, possibly in the context of variable androgen receptor (AR) signaling that changes during a man’s lifetime. Recent in silico analysis by our group on potential association between germline variants and PCa specific somatic lesions identified a non-coding polymorphic regulatory element at the 7p14.3 locus associated with DNA repair and hormone regulated transcript levels and with an early recurrent prostate cancer specific somatic mutation in the Speckle-Type POZ protein (SPOP) gene (OR=5.54, P=1.22e-08) in human prostate tissue data. In order to functionally characterize the polymorphic 7p14.3 locus (rs1376350, single nucleotide polymorphism, G>A), we set up to establish isogenic cell lines harboring the minor allele by using the CRISPR/Cas9 system. In parallel, CRISPR/Cas9 system was used to knock out different portion of the region encompassing the 7p14.3 variant and to eliminate transcription factors (TFs) binding sites that were identified from previous in silico analysis (i.e. AR and CCAAT/Enhancer Binding Protein (C/EBP) beta (CEBPβ)). The transcriptomes of edited pools and edited single clones from macrodeletion (731 bp), microdeletion (50 bp) and alterations of TFs binding sites were analyzed and compared to the transcriptomes of isogenic cells heterozygous (A/G) and homozygous (A/A) for the minor allele A of the risk variant rs1376350 (with or without AR overexpression). These data identified a set of genes scattered throughout the genome with the same pattern of deregulation suggesting the implication of the variant on the regulation of genes residing in different chromosomes. Additionally, ChIP-qPCR experiments for histone modification supported the identification of the 7p14.3 locus with enhancer activity. Furthermore, ChIP-qPCR of histone mark associated with transcriptional activation or repression in isogenic cells harboring the minor allele A upon AR overexpression showed that the activity of the locus is higher for the minor allele A compared to G, independently from AR activation. Despite the limitations of our model and the current lack of validation in other cells, we confirmed that some of the differentially expressed genes that emerged from the comparative analysis of edited cells are deregulated in human normal and tumor prostate samples as well. This work is a proof of concept of germline predisposition to molecularly distinct cancer subclasses and has the potential to nominate new mechanisms of cancer development. Future work aims to elucidate the mechanisms implicated in the deregulation of the transcriptome by combining the information obtained until now with potential new players that we expect to identify by Mass Spectrometry experiments. To clarify the link between the 7p14.3 variant and the somatic mutations in SPOP, we plan to express mutant SPOP in isogenic cells harboring the minor allele and to asses DNA damage response upon overexpression or silencing of TFs binding at and around the rs1376350 variant. My work is an example of how the CRISPR/Cas9 system can be used to develop a technical framework with convergent approaches to functionally characterize polymorphic regulatory regions including but not limited to the establishment of isogenic cells upon single nucleotide editing.

CRISPR-Cas is the state-of-the-art biological tool that allows precise and fast manipulation of the genetic information of cellular genomes. The translation of the CRISPR-Cas technology from in vitro studies into clinical applications highlighted a variety of limitations: the currently available systems are limited by their off-target activity, the availability of a Cas-specific PAM sequence next to the target and the size of the Cas protein. In particular, despite high levels of activity, the size of the CRISPR-SpCas9 editing machinery is not compatible with an all-in-one AAV delivery system and the genomic sequences that can be targeted are limited by the 3-NGG PAM-dependency of the SpCas9 protein. To further expand the CRISPR tools repertoire we turned to metagenomic data of the human microbiome to search for uncharacterized CRISPR-Cas9 systems and we identified a set of novel small Cas9 orthologs derived from the analysis of reconstructed bacterial metagenomes. In this thesis study, ten candidates were chosen according to their size (less than 1100aa). The PAM preference of all the ten orthologs was established exploiting a bacterial-based and an in vitro platform. We demonstrated that three of them are active nucleases in human cells and two out of the three showed robust editing levels at endogenous loci, outperforming SpCas9 at particular targets. We expect these new variants to be very useful in expanding the available genome editing tools both in vitro and in vivo. Knock-out-based Cas9 applications are very efficient but many times a precise control of the repair outcome through HDR-mediated gene targeting is required. To address this issue, we also developed an MS2-based reporter platform to measure the frequency of HDR events and evaluate novel HDR-modulating factors. The platform was validated and could allow the screening of libraries of proteins to assess their influence on the HDR pathway.

Automatic Post-Editing (APE) aims to correct systematic errors in a machine translated text. This is primarily useful when the machine translation (MT) system is not accessible for improvement, leaving APE as a viable option to improve translation quality as a downstream task - which is the focus of this thesis. This field has received less attention compared to MT due to several reasons, which include: the limited availability of data to perform a sound research, contrasting views reported by different researchers about the effectiveness of APE, and limited attention from the industry to use APE in current production pipelines. In this thesis, we perform a thorough investigation of APE as a down- stream task in order to: i) understand its potential to improve translation quality; ii) advance the core technology - starting from classical methods to recent deep-learning based solutions; iii) cope with limited and sparse data; iv) better leverage multiple input sources; v) mitigate the task-specific problem of over-correction; vi) enhance neural decoding to leverage external knowledge; and vii) establish an online learning framework to handle data diversity in real-time. All the above contributions are discussed across several chapters, and most of them are evaluated in the APE shared task organized each year at the Conference on Machine Translation. Our efforts in improving the technology resulted in the best system at the 2017 APE shared task, and our work on online learning received a distinguished paper award at the Italian Conference on Computational Linguistics. Overall, outcomes and findings of our work have boost interest among researchers and attracted industries to examine this technology to solve real-word problems.

付記する学位プログラム名: 京都大学卓越大学院プログラム「メディカルイノベーション大学院プログラム」 / 京都大学 / 新制・課程博士 / 博士(医科学) / 甲第25007号 / 医科博第154号 / 新制||医科||10(附属図書館) / 京都大学大学院医学研究科医科学専攻 / (主査)教授 遊佐 宏介, 教授 萩原 正敏, 教授 齋藤 潤 / 学位規則第4条第1項該当 / Doctor of Medical Science / Kyoto University / DFAM

Creation of visually-pleasing images has always been one of the main goals of computer graphics. Two important components are necessary to achieve this goal --- artists who design visual aspects of an image (such as materials or lighting) and sophisticated algorithms that render the image. Traditionally, rendering has been of greater interest to researchers, while the design part has always been deemed as secondary. This has led to many inefficiencies, as artists, in order to create a stunning image, are often forced to resort to the traditional, creativity-baring, pipelines consisting of repeated rendering and parameter tweaking. Our work shifts the attention away from the rendering problem and focuses on the design. We propose to combine non-physical editing with real-time feedback and provide artists with efficient ways of designing complex visual aspects such as global illumination or all-frequency shadows. We conform to existing pipelines by inserting our editing components into existing stages, hereby making editing of visual aspects an inherent part of the design process. Many of the examples showed in this work have been, until now, extremely hard to achieve. The non-physical aspect of our work enables artists to express themselves in more creative ways, not limited by the physical parameters of current renderers. Real-time feedback allows artists to immediately see the effects of applied modifications and compatibility with existing workflows enables easy integration of our algorithms into production pipelines.

No description available.

No description available.