ABSTRACT

Secondary structure prediction approaches rely typically on models of equilibrium free energies that are themselves based on in vitro physical chemistry. Recent transcriptome-wide experiments of in vivo RNA structure based on SHAPE-MaP experiments provide important information that may make it possible to extend current in vitro-based RNA folding models in order to improve the accuracy of computational RNA folding simulations with respect to the experimentally measured in vivo RNA secondary structure. Here we present a machine learning approach that utilizes RNA secondary structure prediction results and nucleotide sequence in order to predict in vivo SHAPE scores. We show that this approach has a higher Pearson correlation coefficient with experimental SHAPE scores than thermodynamic folding. This could be an important step towards augmenting experimental results with computational predictions and help with RNA secondary structure predictions that inherently take in-vivo folding properties into account.

 抽象的

二级结构预测方法通常依赖于平衡自由能模型，该模型本身基于体外物理化学。最近基于 SHAPE-MaP 实验的体内 RNA 结构的转录组范围实验提供了重要信息，这些信息可能使扩展当前基于体外的 RNA 折叠模型成为可能，从而提高计算 RNA 折叠模拟相对于实验的准确性。测量体内RNA二级结构。在这里，我们提出了一种机器学习方法，利用 RNA 二级结构预测结果和核苷酸序列来预测体内 SHAPE 分数。我们表明，与热力学折叠相比，这种方法与实验 SHAPE 分数具有更高的皮尔逊相关系数。这可能是通过计算预测增强实验结果的重要一步，并有助于本质上考虑体内折叠特性的 RNA 二级结构预测。