Molecular and computational tools for regulatory genomics

基因组DNA的转录是调控每个基因表达多少的中心步骤，对发育、细胞特性和对刺激的反应是不可或缺的。细胞使用称为转录因子的蛋白质，这些转录因子以复杂的方式共同工作，解释DNA序列，并在一个被称为“顺式调控逻辑”的过程中调节转录。虽然我们可以使用参考人类基因组来研究顺式调控，但它可能没有包含足够的很少使用的调控的例子来学习。然而，我们可以通过创造新的例子来学习任意复杂的机制。我的研究计划的长期目标是仅从DNA序列预测任何细胞类型的基因表达。在短期内，我将学习使用下一代实验和计算工具选择细胞类型的顺式调控逻辑。我们之前已经证明，随机DNA是学习顺式调控逻辑的理想训练数据。随机DNA可以合成并以非常高的吞吐量进行检测。每个实验都测量了整个人类基因组的调控DNA。随机DNA的活性与实际的调控DNA相当，这是因为随机DNA中含有丰富的顺式调控元件。这些数据的规模和多样性使我们能够从头开始了解复杂的监管机制，极大地促进了我们对顺式监管的理解。我们假设随机DNA将对人类顺式调控产生类似的见解。每个随机序列中的非调控元件的分类将提供丰富的数据，从中了解活跃在每种细胞类型中的转录因子的作用。(1)我们将学习受控系统中复杂的顺式调控机制。我们将开发实验工具，在受控系统和许多细胞类型中测量每个实验数千万DNA序列的调节活性，为我们提供学习复杂调节机制所需的训练数据的规模。(2)我们将使我们的模型适应基因组。通过随机化基因组DNA和测量基因表达，我们将了解顺式调控序列如何影响内源基因的表达。这将使我们能够从全基因组的序列中预测基因表达。(3)学习更复杂的规则。当我们通过前两个目标生成数据时，我们将开发我们的模型来捕获更复杂和罕见的基因调控规则。我们在目标1和2中创建的方法将作为基因调控的哨兵，提供转录细胞状态的直接读数。我们将使我们的计算和实验工具可供该领域的其他人使用，从而最大限度地发挥我们的影响。作为一个混合计算和实验的实验室，这一应用中包含的9个HQP将接受机器学习、大数据、统计学、基因组编辑、合成生物学、分子生物学、基因调控、项目管理和通信方面的培训。

Transcription of genomic DNA is a central step in regulating how much each gene is expressed, and integral to development, cellular identity, and response to stimuli. Cells use proteins called transcription factors that work together in complex ways to interpret DNA sequences and regulate transcription in a process termed "cis-regulatory logic". Although we can study cis-regulation using the reference human genome, it likely doesn't contain sufficient examples of the rarely-used regulatory from which to learn. However, we can learn arbitrarily complex mechanisms by creating new examples. The long-term objective of my research program is to predict gene expression in any cell type from DNA sequence alone. In the short term, I will learn cis-regulatory logic in select cell types with the next generation of experimental and computational tools. We previously showed that random DNA makes ideal training data for learning cis-regulatory logic. Random DNA can be synthesized and assayed in very high throughput. Each experiment measured an entire human genome's worth of regulatory DNA. Random DNA was comparable in activity to actual regulatory DNA due to the abundant cis-regulatory elements included by chance. The scale and diversity of these data allowed us to learn complex regulatory mechanisms from scratch, significantly advancing our understanding of cis-regulation. We hypothesize that random DNA will yield similar insights into human cis-regulation. The assortment of  -regulatory elements in each random sequence will provide rich data from which to learn the roles of the transcription factors active in each cell type. (1) We will learn complex cis-regulatory mechanisms in a controlled system. We will develop experimental tools for measuring the regulatory activity of tens of millions of DNA sequences per experiment in a controlled system and in many cell types, giving us the scale of training data needed to learn complex regulatory mechanisms. (2) We will adapt our models to the genome. By randomizing genomic DNA and measuring gene expression, we will learn how cis-regulatory sequences affect the expression of endogenous genes. This will allow us to predict gene expression from sequence genome-wide. (3) Learning more complex rules. As we generate data through the first two objectives, we will develop our models to capture more complex and rare gene regulatory rules. The approaches we create in Objectives 1 and 2 will act as gene regulation sentinels, providing a direct readout of the transcriptional cell state. We will make our computational and experimental tools accessible to others in the field, maximizing our impact. As a mixed computational and experimental lab, the 9 HQP included in this application will be trained in machine learning, "big data", statistics, genome editing, synthetic biology, molecular biology, gene regulation, project management, and communication.