SgRNA Activity Prediction Method Based On Reinforcement Learning

Gene-editing technology has great potential in biomedical research,gene breeding and drug development.CRISPR/Cas9 is the most widely used gene-editing technology at present.However,CRISPR gene editing relies on efficient sg RNA design,so developing methods to predict sg RNA activity with high accuracy is crucial.At present,deep learning is the primary method for sg RNA activity prediction,and the prediction of sg RNA activity by deep learning method has attracted extensive attention from researchers.However,most existing deep learning models are based on manual design models,which are cumbersome and time-consuming.Reinforcement learning has the ability of self-learning,so this paper uses it to generate a prediction model of sg RNA activity.In this paper,we use deep learning,reinforcement learning,and neural architecture search to develop an algorithm called Reinforce CRISPR,and we use Reinforce CRISPR to generate GenCRISPR model,which is used in the sg RNA activity prediction of CRISPR/Cas9 system.At last,we develop a web application to predict sg RNA activity.The main content of this paper is as follows:(1)This paper proposes a new method called Reinforce CRISPR,which makes use of RNN controller to generate sg RNA activity prediction model,and uses dataset training feedback to improve the automatically generated model.This method can save the time of model design.(2)The GenCRISPR model is generated by using the Reinforce CRISPR Algorithm.The average Spearman coefficient,Pearson coefficient,and AUC of this model in six data sets are 0.864,0.873 and 0.925 respectively.By comparing Spearman coefficient,Pearson coefficient,and AUC.Experiments show the GenCRISPR model significantly outperform several state-of-the-art algorithms.(3)Based on the GenCRISPR model,we developed a tool for sg RNA activity prediction,The tool can read the base pair uploaded by the user,output the activity of the sg RNA.Furthermore,it can generate a model based on the dataset.