Developing parallel ant colonies filtered by deep learned constrains for predicting RNA secondary structure with pseudo-knots

摘要

RNA plays important roles in cells besides being a simple carrier of genetic information. RNA secondary structure prediction is an efficient way to explore its biochemical function. RNA secondary structure prediction with pseudo-knots is really difficult, which has been proven as an NP-hard problem. Many of the existing predictions seemed to be limited not only by the quality of the search algorithms but also by the quality of the objective functions used. In this paper, a novel prediction model called DpacoRNA is proposed to improve the accuracy of the RNA secondary structure prediction with pseudo-knots, which mainly consists of two features: a parallel search algorithm and learning structural constraints using a deep model. First, based on the base-paired and single-stranded probabilities as multiple objective functions, DpacoRNA applied a parallel ant colony optimization strategy to predict the RNA secondary structure. Second, a bi-directional LSTM recurrent neural network was used to learn the base-pairing constraints. Finally, the constraints learned from the deep model were applied to the output of parallel ant colonies to refine the final secondary structures. To examine the strength and the weakness of the proposed method, multiple RNA types, including RNase P RNA, 5s rRNA, hammerhead ribozyme, transfer RNA, and tmRNA, were used to carefully benchmark DpacoRNA with other state-of-the-art solutions. The final results showed that DpacoRNA was competitive to the other methods. (C) 2019 Elsevier B.V. All rights reserved.