RNA二级结构和RNA结合蛋白相互作用的动态关系和调控

In different types of cells and also in the course from being transcribed to transferred to different cellular compartments for its functioning, RNA interacts with many other different molecules including RNA binding proteins (RBPs), a process often coupled with RNA structure changes. As research suggested, RNA structures play very important roles in determining RBP-RNA interactions. However, due to the lack of in vivo RNA structure information, the knowledge of RNA structure dynamics and its interplay with RBP-RNA interactions remains very limited. .The objective of this proposal is to establish a machine-learning model of RBP-RNA interactions that include RNA structure information. We first use icSHAPE, our newly established transcriptome-wide in vivo RNA secondary structure probing technology, to obtain structure information for RNAs of different types of cells and from different subcellular compartments. We then use deep learning algorithms to build a convolutional neural network of RBP-RNA interaction model for further validation, based on our RNA structure probing data and RBP-RNA interaction data retrieved from public available databases of CLIP-seq experiments. Once trained in one cellular condition, this neural network model can predict RBP-RNA interactions in another condition as long as RNA structure information is available. In particular, genomic variations can affect RBP-RNA interaction by changing RNA structures, and finally possibly result in diseases. Thus this model can be used to predict the disease risk of certain mutations and find its applications in the practice of precision medicine.

在不同类型细胞内，以及从被转录到被转运到细胞内其他位置发挥功能的过程中，RNA会与其他不同分子包括RNA结合蛋白（RBP）结合，其结构也会发生动态变化。RBP-RNA相互作用与RNA结构紧密相关。然而由于缺乏细胞内RNA结构信息，我们对这两者之间的相互影响了解不多。.本项目通过机器学习建立包含RNA结构信息的RBP-RNA相互作用模型。首先利用一种我们新建立的、可以在细胞内测得RNA二级结构信息的技术icSHAPE，来获得不同细胞系不同亚细胞组分的RNA结构数据。进一步利用深度学习，结合icSHAPE RNA结构数据和数据库中收集的RBP-RNA结合位点数据，建立RBP-RNA相互作用模型并进行验证。该模型能够预测更多条件下RBP-RNA的结合。特别的，RNA序列的变化可以影响其结构进而影响RBP结合并引发疾病，因而该模型可以用来预测基因变异的患病风险，从而在精准医疗方面得到应用。

RNA 结构可以影响基因表达调控的各个阶段。现在的技术可以解析全转录组的RNA二级结构信息。但是，之前的RNA结构信息都是整个细胞一起解析结构，不能检测到细胞不同组分的RNA结构变化。我们通过亚细胞分离，解析了染色质，细胞核，细胞质在人以及鼠细胞中的RNA结构。不同组分的RNA结构拓展了RNA结构的信息，更精细的描绘了RNA结构在基因转录，翻译以及降解过程中的核心作用。另外，我们通过研究RNA结构与RBP-RNA相互作用，预测并验证了新的RNA修饰阅读蛋白，比如LIN28A。进一步，我们利用RNA结构信息开发了基于深度学习算法的RBP-RNA相互作用模型PrismNet。我们构建了168个RNA结合蛋白的模型，并利用关注策略发现了蛋白识别RNA的准确的序列与结构信息。通过模型，我们发现引起RNA结构变化的突变富集在RBP结合的区域并且会富集在人类疾病位点，这为我们方法在精准医疗方面的应用奠定了基础。