Unleashing the Genetic Architecture of Heritable Traits | Christoph Lippert скачать в хорошем качестве

Unleashing the Genetic Architecture of Heritable Traits | Christoph Lippert

Abstract This presentation highlights the integration of genomics, imaging, and machine learning in unraveling the intricate genetic architecture of heritable traits. The discussion commences by showcasing the fusion of deep learning and statistical hypothesis testing, harnessing the power of deep neural network representations to enhance statistical analyses of images. Building upon this foundation, we introduce transferGWAS, an approach that directly applies deep learning to conduct genome-wide association studies of medical images. Our exemplar study on retinal fundus images uncovers novel candidate loci associated with eye-related traits and diseases. Furthermore, we delve into the intricacies of an exome-wide association study, where deep-learning-based functional annotations of the genome enable kernel-based tests to identify significant gene-biomarker associations, facilitating a more interpretable understanding of genetic determinants. Expanding the horizons, we introduce ContIG, our cutting-edge self-supervised multimodal contrastive learning method, enabling the exploration of vast datasets comprising unlabeled medical images and genetic data to unveil cross-modal associations. Finally, we provide insights into our ongoing research within the INTERVENE consortium, illustrating the integration of these innovative methods toward the development of effective polygenic risk scores. Collectively, these endeavors aid our comprehension of the genetic foundations of heritable traits and open new avenues for disease risk prediction. https://www.dkfz.de/en/datascience/se...