Research On RNA Secondary Structure Prediction

In the past few years,scientists have found that many DNA fragments are transcribed into RNA,but these noncoding RNAs are not translated into proteins,they have specific structures.These RNAs can perform specific functions such as catalytic reactions,regulation of gene expression,and the like.The general functions of RNA are related to their structures,which also makes people continue to study the structure of RNA.The RNA structure can be measured by experiments,and there are two main methods,nuclear magnetic resonance and X-ray crystal diffraction.However,the method of experimental measurement is difficult to produce samples,and the cost is relatively high,which is time consuming and laborious.If the secondary structure of RNA can be obtained by computer simulation,it will save a lot of resources.Therefore,using computational methods to obtain the structure of RNA is a research hotspot.With the growth of a large amount of sequence information,and the development of computer science,theoretical predictions has become more reliable.The function of non-coding RNA is usually related to its tertiary structure.In order to obtain the tertiary structure of RNA,the secondary structure prediction of RNA is an important link.RNA secondary structure prediction mainly includes two methods,based on minimum free energy and co-evolution information,each of which has advantages and disadvantages.Recently,machine learning and deep learning have achieved good results in many fields.For example,AlphaGo has defeated human Go players,and has performed well in image processing,natural language processing,protein contact prediction,and drug screening.Recently,AlphaFold has also achieved good results in the field of protein structure prediction.This article mainly carried out the following work,firstly the neural network method is used to predict the pairing state of the residues,and then the paired state information of the residues is used as a constraint to remove some of the false positives of direct coupling analysis,but the effect is not good;then the fully convolutional neural network was used and two neural network models were trained to predict the secondary structure of RNA,and then the results were optimized by base extension.In the first model,the result of direct coupling analysis is used as input,and the accuracy is improved.In the second model,the statistical result of multiple sequence alignment is used as input,which is better than the first model and other popular methods in most cases.