Computational Approaches for RNA Structure and Function Determination

在与Shuo Gu的合作中，我们使用3种不同的处理方法（天然细胞裂解物、去除蛋白质、去除核糖体RNA和蛋白质），对4种不同细胞系的转录组RNA-RNA相互作用（长度超过18个核苷酸的RNA双链）进行了全面的研究。用靶向单链rna的核糖核酸内切酶处理后检测MySeq读数。这些读数与各种潜在相互作用的计算生物信息学分析相关。确定其流行与否是为了更好地了解细胞RNA如何与其细胞环境相互作用。有趣的是，主要发现是在4个细胞系中发现的RNA-RNA相互作用很少，因此表明这种相互作用是避免的。当从裂解物中去除蛋白质并重新退火序列时，相互作用的数量显着增加。大多数双链涉及核糖体转录物。来自不同细胞系的不同结果表明，RNA-RNA相互作用在很大程度上是随机的。细胞可能会避免这种相互作用，以防止激活固有免疫反应，这通常是为病毒保留的。Drosha在细胞系统中的功能对于理解microrna的加工及其与正常细胞活动以及癌症等疾病的关系非常重要。在与Shuo Gu的另一项合作中，研究了Drosha靶向茎环结构与产生的microRNA异构体类型的关系。采用实验和计算方法来确定这些关系。结果表明，目标Drosha茎中的弯曲、扭曲和/或柔性结构似乎促进了microRNA替代形式的产生。对结构预测结果与实验结果进行了比较和关联。具体来说，对pri-miR-9-1的切割，而不是对pri-miR-9-2或pri-miR-9-3的切割，会产生具有移位种子序列的替代miR-9，从而扩大其靶rna的范围。有趣的是，对低级别胶质瘤患者样本的分析表明，替代的pri-miR-9在肿瘤进展中具有潜在的作用。-与Esta Sterneck实验室的合作正在进行中。她的实验室利用体外细胞培养和体内小鼠模型系统研究了参与乳腺和胶质母细胞瘤肿瘤发生的细胞信号通路，重点研究了转录因子CCAAT/增强子结合蛋白δ (CEBPD)。利用转基因乳腺癌小鼠模型，她的研究小组发现CEBPD在乳腺肿瘤发生中具有双重作用。一方面，CEBPD抑制肿瘤的多样性，另一方面，CEBPD促进远处肺转移。此外，CEBPD通过调节各种信号通路和干性，促进乳腺癌和胶质母细胞瘤细胞中的干细胞样癌细胞，干细胞样癌细胞与肿瘤转移和治疗耐药有关。此外，为了下调CEBPD介导的肿瘤进展信号，需要靶向CEBPD信息的策略。作为我们纳米生物学项目的一个结合，我们的实验室正在开发RNAi疗法的方法，通过递送策略设计的RNA纳米结构作为它们自己的实体或与脂质载体结合来敲除CEBPD mRNA。我们基于脂质载体的初步结果看起来很有希望，我们目前正在使用小鼠模型进行进一步验证。细胞形状预测为确定细胞内RNA结构提供了一个新的细节水平。这些结果可能与更标准的SHAPE技术不同，后者在产生潜在的结构预测时不考虑细胞环境。我们正在开发一种基于大肠杆菌数据库中获得的已知细胞形状测量数据，通过训练神经网络（已通过超参数化技术进行优化）来计算预测细胞形状的方法。在给定序列的情况下，预测似乎能产生相当准确的结果。这是首次将计算方法应用于基于机器学习的细胞内RNA结构预测。——预测含有非规范碱基对相互作用的RNA二级和三维结构是一个困难而重要的问题，需要更好的算法。我们正在开发一套基于贝叶斯和神经网络方法的计算算法，以预测RNA中的规范和更重要的非规范碱基对相互作用。已经编译了一个包含大量结构的大型数据库，其中包括非规范碱基对相互作用。这些算法已经显示出显著的实用性，可以在二级结构水平上预测复杂的基序。然后将这些结果与RNA 3D结构生成程序结合使用，该程序可以预测包含复杂非规范相互作用的3D RNA结构。这套算法也被应用于多序列RNA纳米组装的预测。—在与Mikhail Kashlev合作的另一个项目中，我们正在开发一种使用神经网络方法（机器学习）来确定细菌细胞中的转录暂停位点的方法。序列读取是由一个叫做RNet-seq （Net-seq的变体）的方法产生的，用来定义暂停点。利用这些reads开发的神经网络可以在不同的细菌序列（大肠杆菌和枯草芽孢杆菌）中区分这些类型的位点。该方法似乎表明，这些细菌系统在暂停方面存在显着差异。除了辅助因素外，暂停位点周围的序列和结构变化也起着重要作用。所有这些信息都被整合到机器学习方法中，以帮助识别过程。-与Stuart Le Grice合作的另一个项目涉及开发一种计算方法来确定针对各种RNA结构基序的小分子的结合位点和亲和力。这个项目的目标是帮助筛选小分子，因为它们有可能在治疗上对靶向病毒rna或致癌基因有益。这些小分子最初是通过结合小分子微阵列的实验筛选方法得到的。管道，因为它目前的立场是能够确定一个合理的精度水平配体的姿势，以及结合口袋的构象。它似乎还能区分不同配体的不同结合亲和力。一种名为RiboSketch的算法已经被开发出来，用于描述核酸二级结构，这些结构可能包含由RNA或DNA组成的多条链。布局算法，全面的编辑能力，并在系统中提供多种仿真模式。交互功能允许RiboSketch创建具有广泛组成，大小和复杂性的结构的出版质量图表。该程序可以在任何web浏览器中运行而无需安装，也可以作为独立的Java应用程序运行。

In collaboration with Shuo Gu, we have pursued a comprehensive examination of transcriptome-wide RNA-RNA interactions (RNA duplexes longer than 18 nucleotides) across 4 different cell lines using 3 different treatments (native cell lysates, removal of proteins, and removal of ribosomal RNA and proteins). MySeq reads were examined after treating with endoribonucleases targeting single stranded RNAs. These reads were correlated with a variety of computational bioinformatic analyses of the potential interactions. The prevalence or lack thereof was determined to enable a better understanding of how cellular RNA interacts with its cellular environment. Interestingly the major finding was that there are very few RNA-RNA interactions found across the 4 cell lines, thus indicating that such interactions are avoided. The number of interactions went up significantly when proteins were removed from the lysates and the sequences re-annealed. The majority of duplexes involved ribosomal transcripts. The diversity of results from the different cell lines suggests that RNA-RNA interactions are to a large extent stochastic in nature. Cells presumably avoid such interactions to prevent activation of innate immune responses, which are normally reserved for viruses. ---The functionality of Drosha in cellular systems is important for understanding the processing of microRNAs and how they relate to normal cellular activity as well as diseases such as cancer. In another collaboration with Shuo Gu the relationship of Drosha targeted stem-loop structures and the type of microRNA isforms that are produced was examined. Experimental and computational approaches were applied to determine these relationships. Results indicate that bent, distorted and/or flexible structures in the targeted Drosha stem seem to facilitate the production of alternate forms of microRNA. Structural predictions and experimental results were compared and correlated. Specifically, cleavage of pri-miR-9-1, but not pri-miR-9-2 or pri-miR-9-3, generates an alternative miR-9 with a shifted seed sequence that exapands the scope of its target RNAs. Interestingly, analysis of low-grade glioma patient samples indicate that alternative pri-miR-9 has a potential rolein tumor progression. ---A collaboration with Esta Sterneck's laboratory is ongoing. Her lab investigates cell signaling pathways involved in breast and glioblastoma tumorigenesis with a focus on the transcription factor CCAAT/enhancer binding protein delta (CEBPD) using in vitro cell culture and in vivo mouse model systems. Using a transgenic mouse model of breast cancer, her group has shown that CEBPD exhibits a dual role in mammary tumorigenesis. On the one hand, CEBPD prevents tumor multiplicity and on the other hand, CEBPD promotes distant lung metastases. In addition, CEBPD promotes stem-like cancer cells, which have been implicated in tumor metastasis and treatment resistance, in breast and glioblastoma tumor cells through regulation of various signaling pathways and stemness. In addition, strategies for targeting the message of CEBPD are necessary to down regulate CEBPD-mediated tumor progression signaling. As a tie-in to our nanobiology project, our laboratory is developing approaches for RNAi therapeutics to knock down the CEBPD mRNA by delivering strategically designed RNA nanostructures as their own entities or in combination with lipid carriers.Initial results with our lipid-based carriers look promising and we are currently progressing to the use of mouse models for further verification. --- In cell SHAPE prediction provides a new level of detail for determining RNA structure within cells. These results may vary from the more standard SHAPE techniques that do not take the cellular environment into account when producing potential structural predictions. We are developing a method for the computational prediction of in cell SHAPE by training a neural network (which has be optimized by hyper-parmaterization techniques) based on known in cell SHAPE measurements obtained from an E. coli database. Predictions, given a sequence, seem to be producing reasonably accurate results. This is the first time that computational methods have been applied to the prediction of RNA structure within cells based on machine learning. ---The prediction of RNA secondary and 3D structures containing non-canonical base pair interactions is a difficult and important problem that needs better algorithms. We are developing a set of computational algorithms based on Bayesian and neural network methodologies to enable the prediction of canonical and more importantly non-canonical base pair interactions in RNA. A large database has been compiled containing a multitude of structures including the non-canonical base pair interactions. The algorithms have shown significant utility, enabling the prediction of complex motifs at the secondary structure level. These results are then being used in conjunction with an RNA 3D structure generation program, which enables the prediction of 3D RNA structures that incorporate the complex non-canonical interactions. This set of algorithms are also being applied to the prediction of multi-sequence RNA nano-assemblies. ---In another project in collaboration with Mikhail Kashlev we are developing a methodology using a neural network approach(machine learning) to determine transcriptional pause sites in bacterial cells. Sequence reads are being produced by a method called RNet-seq (a variation of Net-seq) to define pause sites. The neural net being developed using these reads discriminates between these types of sites in various bacterial sequences (E. coli and B. subtilis). The approach seems to indicate that their are significant differences between these bacterial systems in regards go pausing. Sequence and structural variations around the pause site play a significant role in addition to co-factors. All this information in being incorporated into the machine learning approaches to aid in the discrimination process.---Another project in collaboration with Stuart Le Grice involves the development of a computational approach to determined binding sites and affinities of small molecules targeting various RNA structural motifs. The goal of this project is to aid in the screening of small molecules for their potential to be therapeutically beneficial in targeting viral RNAs or cancer causing genes. The small molecules are initially derived from sets found by binding to experimental screening methods using small molecule microarrays. The pipeline as it currently stands is able to determine to a reasonable level of accuracy ligand poses as well as the conformation of the binding pockets. It also seem able to discriminate between different levels of binding affinities for different ligands. ---An algorithm, RiboSketch, has been developed for the depiction of nucleic acid secondary structures which may contain multiple strands consisting of RNA or DNA. Layout algorithms, comprehensive editing capabilities, and a multitude of simulation modes are available within the system. Interactive features allow RiboSketch to create publication quality diagrams for structures with a wide range of composition, size, and complexity. The program may be run in any web browser without the need for installation, or as a standalone Java application.