Regulatory models for RNA processing

在过去的5年里，高通量生物学数据集的数量呈指数级增长，这使得我们有可能想象一个真正的定量生物学和数据驱动医学的时代。然而，更多的数据并不一定会带来更好的理解或深刻的见解，对于生物数据来说尤其如此。与数据分析基因组、生物分子和细胞过程的指数级增长相比，我们在知识和医学应用方面的收获非常有限。最近，人工智能技术，特别是机器学习（ML）和“深度学习”技术，已被证明在利用大数据集方面是有效的。深度学习已经彻底改变了计算机视觉和自然语言处理等应用领域。作为一个研究基因组生物学和机器学习的实验室，我们的目标是开发充分利用大型生物数据集和推进基因组研究的技术。我的团队在过去几年中取得了重大进展，发表了两篇极具影响力的论文（Xiong et al, Science 2015, Alipanahi et al, Nature Biotechnology 2015）和一家初创公司Deep Genomics。******人体由数万亿个细胞组成，这些细胞的过程控制着正常的功能和疾病状态。然而，所有的细胞都含有相同的遗传信息，即由DNA分子组成的基因组。细胞之间的差异主要是由基因组中的“调节”指令控制的，这些指令指示何时、何地以及产生每种细胞分子的数量，无论是DNA、RNA还是蛋白质。我的长期研究目标是引入有原则的模型和ML算法，这些模型和算法可以解开监管的决定因素，并推断出促进科学理解和实现生物医学技术的监管模型。在本提案中，我们将重点关注三个不同但相关的生化过程：蛋白质与DNA和RNA分子的结合，选择性聚腺苷化和RNA编辑，并使用深度学习技术建立准确的调节模型。这些过程都是细胞调节网络的重要参与者，它们的破坏有助于广泛的人类疾病，如神经系统疾病和癌症。我相信我们可以在这个具有挑战性的项目中取得成功，基于我们小组在基因组生物学和ML方面培养的多样化专业知识，以及我们之前在监管模型方面的工作中获得的经验和教训，以及有希望的初步结果。我希望这项工作能够在基因组生物学方面取得重大进展，并实现一系列下游应用，例如基因检测和治疗发现。它还将有助于培养该领域的世界级领导者以及科学家和工程师，这将使社区和加拿大受益。

Over the past 5 years, there has been exponential growth in the number of high-throughput biology datasets, making it possible to imagine an era of truly quantitative biology and data-driven medicine. However, more data doesn't necessarily lead to better understanding or deep insights, and this is especially true for biological data. Compared to the exponential growth of data profiling genomes, biomolecules and cellular processes, our gain in knowledge and applications to medicine are very limited. Recently, artificial intelligence techniques, especially techniques for machine learning (ML) and “deep learning”, have proven to be effective in taking advantage of big datasets. Deep learning has revolutionized application areas such as computer vision and natural language processing. As a research lab working on genome biology and ML, our goal is to develop techniques to make full use of large biological datasets and advance genomic research. My group has made significant progress in the past few years, resulting in two highly influential publications (Xiong et al, Science 2015, Alipanahi et al, Nature Biotechnology 2015) and a startup spin-out, Deep Genomics. ******The human body consists of trillions of cells whose processes govern normal function and also disease states. Yet, all the cells contain the same genetic information, the genome, which is made up of DNA molecules. Differences between cells are mainly governed by the ‘regulatory' instructions in the genome, which instruct when, where, and how much to produce each kind of cellular molecule, whether DNA, RNA or protein. My long-term research objectives are to introduce principled models and ML algorithms that can disentangle the determinants of regulation, and to infer regulatory models that advance scientific understanding and enable biomedical technologies. In this proposal, we will focus on three distinct, but related, biochemical processes: protein binding to DNA and RNA molecules, alternative polyadenylation and RNA editing, and build accurate regulatory models using deep learning techniques. These processes are all important players of the regulatory network in a cell, and disruption of them contributes to a wide range of human diseases, such as neurological disorders and cancer. I believe we can succeed in this challenging project, based on the diverse expertise cultivated in my group on both genome biology and ML, the experience and lessons learned from our previous work on regulatory models, and promising preliminary results. I expect this work to lead to major advances in genome biology and enable a range of downstream applications, such as in genetic testing and therapy discovery. It will also help to train world-class leaders in the field as well as scientists and engineers, which will benefit the community and Canada.