Deep-learning based profiling of patient-derived cells as a tool for genomic and translational medicine

基于深度学习的患者来源细胞分析作为基因组和转化医学的工具

基本信息

批准号：
10321280
负责人：
Wolfgang Maximilian Anton Pernice
金额：
$ 9.79万
依托单位：
COLUMBIA UNIVERSITY HEALTH SCIENCES
依托单位国家：
美国
项目类别：
财政年份：
2020
资助国家：
美国
起止时间：
2020-12-22 至 2023-04-30
项目状态：
已结题

来源：
https://reporter.nih.gov/project-details/10321280
关键词：
Accounting Address Award Benign Biological Cells Cellular Morphology Cellular biology Clinical Complement Complex Computers Custom Data Data Science Data Sources Development Development Plans Diagnostic Disease Environment Face Fibroblasts Fostering Gene Expression Profiling Genes Genetic Genetic Diseases Genetic Variation Genome Genomic medicine Genomics Goals Hand Heritability Image Institutes Intervention Lead Leadership Machine Learning Malignant Neoplasms Maps Measures Medical Genetics Methods Mitochondrial Diseases Molecular Morphology Nature Nerve Degeneration Outcome Pathogenicity Patient imaging Patients Pharmacology Phenocopy Phenotype Privatization Protocols documentation Rare Diseases Research Research Personnel Resolution Sample Size Scientist Signal Transduction Standardization Techniques Technology Testing Therapeutic Intervention Time Training Universities Validation Variant base career cellular imaging cohort cost efficient deep learning deep learning algorithm design efficacy testing experimental study gene therapy genetic association genome wide association study genome-wide genomic tools hereditary neuropathy individual patient learning strategy microscopic imaging multidisciplinary patient subsets patient variability personalized genomic medicine programs single-cell RNA sequencing skills small molecule libraries statistics success tool transcriptome sequencing translational medicine

项目摘要

Project Summary/Abstract: The genetic landscape of rare and common diseases has emerged as heterogeneous and complex. Already, researchers and clinicians face the challenge to discern pathophysiological mechanism and treatment opportunities for hundreds of genetic subtypes that have been identified in rare diseases, such as inherited neuropathies (INs) or mitochondrial diseases (MiDs) alone. Still, a large fraction of disease loci remains to be discovered – a daunting task, since gene-identification studies often require immense sample-sizes, which are difficult to achieve, even for more common conditions. Simultaneously, much of the heritability of many disorders appears to be determined by the collective impact of possibly thousands of low-impact variants, spread across the genome. Ideally, the impact of a given set of candidate variants could be assessed within high-throughput framework that accounts for the genetic context of individual patients. Leveraging advanced deep learning algorithms, we have developed an unbiased, scalable method to rapidly identify disease- associated phenotypes in high-resolution, multiplexed, fluorescent microscopy images of primary, patient derived cells. In turn, the discovered phenotypes can be exploited as experimental signals against which the disease relevance of candidate variants can be confirmed, by virtue of genetic complementation experiments. At the same time, the standardized and scalable nature of our method renders it suitable to test potential therapeutic interventions, e.g. to test the efficacy of potential gene-therapy, or to screen small molecule libraries, while maintaining patient-specific granularity. The goal of this proposal is to apply our approach to an expanded cohort of patient cells and to refine methods to interpret both genetic and pharmacological perturbations. In this, I will be supported by an exceptional and multidisciplinary team of experts in clinical, molecular and functional genetics, and computer scientists, within the world-class scientific environment offered by Columbia University and the Broad Institute. In a carefully designed development plan, I will finalize my training in machine learning and data science, expand my expertise to single-cell RNA-sequencing and other single-cell methods, and acquire essential leadership and scholarly skills required for an independent research career. Over the course of this award, I will apply our cellular profiling approach to generate a standardized map of deep, quantitative descriptions of disease-associated cellular phenotypes across a number of INs, MiDs and neurodegenerative conditions. We will explore the integration of RNA-sequencing to enhance our approach. Finally, we will apply our method to the discovery and confirmation of new disease genes, and screen a limited number of pharmacological interventions through our method. Together, the proposed developmental plan and research strategy will foster my ability to lead an independent research program, to establish cellular profiling as a powerful platform to advance genomic and translational medicine.

项目摘要/摘要：稀有疾病和常见疾病的遗传景观已成为异质和复杂。已经，研究人员和临床医生面临识别病理生理机制和治疗的挑战在罕见疾病中鉴定出的数百种遗传亚型的机会，例如继承单独使用神经病（INS）或线粒体疾病（MIDS）。尽管如此，很大一部分疾病基因座仍然是发现的 - 一项艰巨的任务，因为基因识别研究通常需要巨大的样本大小，即使在更常见的条件下也很难实现。同时，许多人的遗传力疾病似乎取决于可能数千种低影响变体的集体影响，遍布基因组。理想情况下，可以在其中评估一组给定候选变体的影响高通量框架解释了个别患者的遗传环境。利用高级深度学习算法，我们已经开发了一种公正的可扩展方法来快速识别疾病 - 在高分辨率，多路复用，荧光显微镜图像中的相关表型的主要患者衍生细胞。反过来，发现的表型可以作为实验信号探索通过遗传完成实验，可以证实候选变异的疾病相关性。同时，我们方法的标准化和可扩展的性质使其适合测试潜力治疗干预措施，例如测试潜在基因疗法的效率，或筛选小分子图书馆，同时保持患者特定的粒度。该提议的目的是将我们的方法应用于扩大患者细胞的队列并完善解释遗传和药物的方法扰动。在这方面，我将得到一个临床专家的杰出和多学科团队的支持，世界一流的科学环境中的分子和功能遗传学以及计算机科学家由哥伦比亚大学和广大研究所提供。在精心设计的开发计划中，我将最终确定我在机器学习和数据科学方面的培训，将我的专业知识扩展到单细胞RNA测序和其他单细胞方法，并获得独立的基本领导和科学技能研究职业。在这个奖项的过程中，我将采用我们的蜂窝分析方法来产生跨A的疾病相关细胞表型的深层，定量描述 INS，MIDS和神经退行性条件的数量。我们将探讨RNA顺序的集成到增强我们的方法。最后，我们将把我们的方法应用于新疾病的发现和确认基因和通过我们的方法筛选有限数量的药物干预措施。在一起，拟议的发展计划和研究策略将促进我领导独立研究的能力程序，以建立蜂窝分析作为推进基因组和翻译医学的强大平台。