Defining the protein sequence features that control transcriptional activation domain function

项目摘要/摘要 大多数细胞类型特异性基因的表达来自于结合到 促进者和促进者12，13.在过去的十年中，识别 转录因子结合数据的增强子和编目14，但仍不可能 从基因组序列预测基因表达，很大程度上是因为我们仍然有一个有限的 了解转录激活区，即结合辅活化蛋白的区域。 我们的团队研究了控制蛋白功能的蛋白质序列特征 转录激活域。我们试图了解激活域是如何工作的，如何 它们的进化以及我们如何从蛋白质序列中预测它们。研究如何激活 域起作用时，我们理性地设计突变来测试特定的假设。研究激活 结构域进化，我们调查了现存的同源同源物的多样性，以发现高度多样化的功能 序列。为了从蛋白质序列中预测激活结构域，我们综合了所有这些数据 使用可解释的机械预报器和卷积神经网络。 我们结合了三种方法来研究激活域函数。首先，我们使用高 酵母和人类细胞培养中的吞吐量分析，合理设计了数千个突变 测试关于函数的特定假设，并将这些数据与机器学习相结合。 其次，我们使用生物物理模拟来研究激活区域的突变是如何改变的 这些内在无序区域的三维结构。第三，我们使用高分辨率成像 研究激活结构域如何调节单个转录因子的运动 活细胞核中的分子。这三种方法将揭示氨基酸 控制激活结构域如何控制转录的序列特征。 我们的长期目标是建立一系列预测激活的计算模型 从蛋白质序列中预测每个激活结构域招募的辅活化子，以及 预测激活域如何进化。

Project Summary/Abstract Most cell-type specific gene expression arises from transcription factors binding to enhancers and promoters12,13. The last decade has seen explosive growth in the identification of enhancers and cataloging of transcription factors binding data14, but it is still not possible to predict gene expression from genome sequence, in large part because we still have a limited understanding of transcriptional activation domains, the regions that bind coactivator proteins. Our group studies the protein sequence features that control the function of transcriptional activation domains. We seek to understand how activation domains work, how they evolve and how we can predict them from protein sequence. To study how activation domains work, we rationally design mutations to test specific hypotheses. To study activation domain evolution, we survey the diversity of extant orthologs to find highly diverse functional sequences. To predict activation domains from protein sequence, we integrate all these data with interpretable mechanistic predictors and with convolutional neural networks. We combine three approaches to study activation domain function. First, we use high throughput assays in yeast and human cell culture, rationally designed thousands of mutations to test specific hypotheses about function, and integrate these data with machine learning. Second, we use biophysical simulations to study how mutations in activation domains change the 3D structures of these intrinsically disordered regions. Third, we use high-resolution imaging to study how activation domains modulate the movements of individual transcription factor molecules in the nuclei of living cells. These three approaches will reveal the amino acid sequence features that control how activation domains control transcription. Our long term goal is to build a family of computational models that predict activation domains from protein sequence, predict the coactivators each activation domain recruits, and predict how activation domains evolve.