基于堆积协变信息与最小自由能预测含伪结的RNA二级结构

Predicting RNA Secondary Structures Including Pseudoknots by Covariance with Stacking and Minimum Free Energy

摘要:

RNA伪结预测是RNA研究的一个难点问题。文中提出一种基于堆积协变信息与最小自由能的RNA伪结预测方法。该方法使用已知结构的RNA比对序列(ClustalW比对和结构比对)测试此方法, 侧重考虑相邻碱基对之间相互作用形成的堆积协变信息, 并结合最小自由能方法对碱基配对综合评分, 通过逐步迭代求得含伪结的RNA二级结构。结果表明, 此方法能正确预测伪结, 其平均敏感性和特异性优于参考算法, 并且结构比对的预测性能比ClustalW比对的预测性能更加稳定。文中同时讨论了不同协变信息权重因子对预测性能的影响, 发现权重因子比值在l1: l2=5:1时, 预测性能达到最优。

Abstract:

Prediction of RNA secondary structures including pseudoknots is a difficult topic in RNA field. Current predicting methods usually have relatively low accuracy and high complexity. Considering that the stacking of adjacent base pairs is a common feature of RNA secondary structure, here we present a method for predicting pseudoknots based on covariance with stacking and minimum free energy. A new score scheme, which combined stacked covariance with free energy, was used to assess the evaluation of base pair in our method. Based on this score scheme, we utilized an iterative procedure to compute the optimized RNA secondary structure with minimum score approximately. In each interaction, helix of high covariance and low free energy was selected until the sequences didn’t form helix, so two crossing helixes which were selected from different iterations could form a pseudoknot. We test our method on data sets of ClustalW alignments and structural alignments downloaded from RNA databases. Experimental results show that our method can correctly predict the major portion of pseudoknots. Our method has both higher average sensitivity and specificity than the reference algorithms, and performs much better for structural alignments than for ClustalW alignments. Finally, we discuss the influence on the performance by the factor of covariance weight, and conclude that the best performance is achieved when l1 : l 2=5 : 1.