48 / 2022-09-26 19:19:31

Prediction of Secondary Structure of SARS-COV-2 Spike Protein Based on Multi-feature Fusion

The novel Coronavirus (SARS-COV-2) is not only rapidly spreading, with a long incubation period and strong vitality, but also constantly mutating. Many reasons may be related to the structure of SARS-COV-2 spike protein.In this experiment, the amino acid sequence data of SARS-COV-2 spike protein were downloaded from NCBI database and coded with various features. The experiment first joint convolution and recurrent neural networks are trained to predict eight secondary structures of proteins. The regularization methods with dropout and l2-norm improved the model's prediction performance to some extent but did not achieve the desired effect. Subsequently, the convolutional block of the previous model is improved inspired by convolutional neural network architecture Inception and Resnet. The improved convolution block makes extensive use of multi-scale convolution kernel synchronization to compute features on multiple scales for the sequence. At the same time, residual learning was added to prevent under-fitting, and the improved model performed better than the previous one.The final results showed that the secondary structure of SARS-COV-2 spike protein was mainly (loop or irregular) and helix, which laid a foundation for further understanding the pathogenesis of SARS-CoV-2 and developing a new vaccine.