Prediction of nearest neighbor parameters for folding RNAs with modified nucleotides

天然和合成的RNA在细胞功能、生物技术和医学中发挥着关键作用。RNA折叠成 复杂的结构，往往驱动它们的功能，因此确定RNA的结构是生物学的基础 和生物技术。基于计算热力学的二级结构建模(TSSM)是一种流行的、 低成本、快速的结构预测方法，这使得转录组范围的结构功能成为可能 合成RNA文库的研究和大量基于结构的筛选。然而，最近的证据表明， 转录后化学核苷酸修饰的多样性还对 局部和/或全局结构，最终调节RNA的稳定性、表达或调节功能。是这样的 修饰广泛存在于所有生命领域，代表着一种新的、鲜为人知的基因层 监管，这与疾病有牵连。此外，它们通常被引入rna药物。 作为一种逃避先天免疫反应的手段。综上所述，自然改造的财富和 新型人工核糖核酸的发展，对其机制的日益关注，以及其在RNA药物中的中心地位 强调迫切需要快速和准确地确定具有修饰核苷酸的RNA的结构。 然而，由于缺乏可供估计的参数，TSSM方法不能考虑修改的影响 它们的折叠稳定性。它们依赖于功能丰富的特纳最近邻(NN)热力学模型，该模型 是由从802个昂贵和费力的正则碱导出的294个自由能变化值来参数化的 紫外线熔化实验。鉴于修改的种类繁多且迅速扩大，重复修改是不切实际的 每种类型的这样的实验。该方案的前提是可以学习更多的神经网络参数 从经济实惠、可广泛访问和高吞吐量的替代实验中高效地获得。 具体地说，下一代测序已经将RNA结构探测(SP)大规模地转变为例行公事 平行实验，报告有关局部核苷酸动力学的结构信息。SP被广泛用于 从全基因组研究中获得对RNA结构和功能的洞察，并将TSSM算法限制为 改善他们的预测。然而，与熔融分析不同，RNA折叠稳定性与SP之间的关系 测量是非常不平凡的，因此从SP数据恢复参数的问题是困难的。 这项提议的目标是开发新的算法和软件来估计神经网络参数 高吞吐量SP数据。我们将设计统计推理方法，使信息与折叠保持一致 算法和SP实验，并将它们应用于未修改和修改的RNA的数据，以估计新的 修饰核苷酸的参数。由于SP数据和折叠热力学之间的联系很复杂，并且 此外，还没有探索从SP数据拟合Turner参数的能力，我们将评估 所开发的方法的可行性、准确性、性能和计算效率。验证 这些努力将包括与实验得出的值进行比较，以及评估对保持不变的数据的预测。

Natural and synthetic RNAs play key roles in cellular function, biotechnology, and medicine. RNAs fold into intricate structures, which often drive their functions, thus determining RNA structure is fundamental to biology and biotechnology. Computational thermodynamics-based secondary structure modeling (TSSM) is a popular, low-cost, and rapid approach to structure prediction, which has enabled transcriptome-wide structure-function studies and massive structure-based screens of synthetic RNA libraries. However, recent evidence suggests that a diversity of post-transcriptional chemical nucleotide modifications additionally exert profound impact on local and/or global structure, to ultimately modulate the RNA’s stability, expression, or regulatory function. Such modifications are widespread in all life domains and represent a new and poorly understood layer of gene regulation, which has been implicated in disease. Moreover, they are routinely introduced into RNA medicines as a means of evading the innate immune response. Taken together, the wealth of natural modifications and development of novel artificial ones, the growing interest in their mechanism, and their centrality to RNA medicine underscore a pressing need to determine structures of RNAs with modified nucleotides rapidly and accurately. However, TSSM methods cannot account for the effects of modifications due to a lack of parameters to estimate their folding stabilities. They rely on the feature-rich Turner nearest-neighbor (NN) thermodynamic model, which is parameterized by 294 free-energy change values derived for canonical bases from 802 costly and laborious UV melting experiments. Given the diverse and rapidly expanding pool of modifications, it is impractical to repeat such experiments for each type. The premise of this proposal is that NN parameters can be learned more efficiently from alternative experiments, which are affordable, widely accessible, and high throughput. Specifically, next-generation sequencing has transformed RNA Structure Probing (SP) into a routine massively parallel experiment, which reports structural information about local nucleotide dynamics. SP is widely used to gain insights into RNA structure and function from genome-wide studies and to constrain TSSM algorithms to improve their predictions. However, unlike melting assays, the relationship between RNA folding stability and SP measurements is highly nontrivial, and thus the problem of recovering the parameters from SP data is difficult. The goal of this proposal is to develop novel algorithms and software to estimate NN parameters from high-throughput SP data. We will design statistical inference methods that reconcile information from folding algorithms and SP experiments and apply them to data for unmodified and modified RNAs to estimate new parameters for modified nucleotides. As the link between SP data and folding thermodynamics is complex, and furthermore, the ability to fit the Turner parameters from SP data has not been explored, we will assess the feasibility, accuracy, performance, and computational efficiency of the developed methods. Validation efforts will include comparing to experimentally derived values and evaluating predictions over held-out data.