Protein-RNA Interaction Prediction Study And Visualization Analysis Based On Deep Learning

As an indispensable macromolecule in life activities,the study of protein-RNA interaction prediction has great influence and significance on bioinformatics and related medical fields(such as the development of Covid-19 vaccine).T Traditional methods to detect protein-RNA interaction pairs are time-consuming and expensive.Therefore,the use of efficient and fast calculation methods and comprehensive data feature representation has become a research trend in protein-RNA interaction prediction.At the same time,with the rapid increase in the amount of data,there is a large amount of information in these massive protein-RNA pair data that is difficult to be discovered by humans.Therefore,using efficient visualization methods to analyze these complex data has become a research hotspot.Firstly,in view of the limitations of the existing prediction models in the calculation of multiple biological information and the incomplete representation of characteristic information,a hybrid deep learning model was designed to predict protein-RNA interactions.In this method,three convolutional neural networks were constructed based on the feature data of protein sequence,RNA sequence and RNA structure to extract their abstract features.And using the bidirectional long and short-term memory network to further learn long-term dependencies between protein and RNA data,finally,the trained parameters are fed to the classification layer to predict protein-RNA interactions.Secondly,in view of the lack of data analysis work involved in protein-RNA interaction prediction,a visualization scheme was designed to analyze the prediction of protein-RNA interactions from multiple perspectives.In this scheme,the training process of protein-RNA interaction prediction,the comparison of prediction results and data characteristics were visualized and analyzed to show the changes and interrelationships between the data in the experiment,explaining the importance of structural information,and the impact of different data processing schemes on the prediction.Finally,experiments are carried out on several datasets,and compared with the existing prediction methods,the effectiveness of the proposed prediction algorithm is verified.At the same time,the feasibility of the visualization scheme is verified by analyzing from many angles.