Prediction Method Of SgRNA On-target Activity Based On Transformer Composite Model

CRISPR/Cas9 system is a widely used genome editing tool,which has broad application prospects in disease treatment and gene function regulation.However,how to accurately predict and evaluate the on-target and off-target effects of single guide RNA(sg RNA)is a key issue in CRISPR/Cas9 system design.Obtaining sg RNA with high sensitivity and specificity by computational methods is an important prerequisite for the optimal design of sg RNAs.At present,researchers have proposed many models for sg RNA target prediction.Although these models have achieved remarkable results in terms of prediction ability,there is still room for improvement in feature processing and model architecture.This paper focuses on how to improve the ability of the model to predict sg RNA activity.The main contents and innovations are summarized as follows:(1)A new model Trans Crispr for predicting sg RNA target activity based on Transformer and CNN architecture is proposed.Trans Crispr captures sequence features of sg RNA at multiple levels,and Transformer structures focus on biologically significant sequence fragments while ensuring comprehensive learning of sequence information to improve the predictive power of models.At the same time,the Transformer module is improved and the dynamic residual structure is introduced to effectively prevent information loss and gradient disappearance.(2)Based on Trans Crispr model,sequence feature fusion and biological feature integration are proposed to further enhance the prediction accuracy of the model.In terms of features,the single base and dimer are mixed to encode,and the sequence data is augmented.In terms of model,the learning module of biological features and Trans Crispr model are integrated to obtain the integrated model of multi-feature fusion,which improves the robustness of the model.In addition,the importance of features was analyzed based on Trans Crispr model pairs,and the biological significance of model training was revealed by the algorithm,which provided a reference for gene editing experiments.(3)Based on the proposed prediction model,the sg RNA targeting activity prediction tool is developed,which can realize the prediction of sg RNA activity and biological characteristics under multiple Cas9 cardinal data sets.In this paper,we compare the prediction model with the mainstream methods on the widely used sg RNA dataset.The experimental results show that the proposed prediction model is superior to the mainstream methods,and the effectiveness of the multi-feature fusion method is also verified,which provides a reference for gene editing experiments.