A classification method for compound mass spectrum data based on the Flex-Bootstrap method and Resnet network

Aiming at the low accuracy of mass spectrum data recognition due to the subjectivity of artificial feature extraction and the limitation of a single sample and unbalanced mass spectrum data in current research, a classification method based on the fusion of the Flex-Bootstrap and Residual Network models is proposed and compared with the DNN model and the fusion model of CNN and DNN. The Flex-Bootstrap method and Resnet fusion model used the Flex-Bootstrap method to expand the combined mass spectrum data perform sparse sequence representation and use the extracted feature vectors to construct the Resnet model. Compared with the traditional classification methods, the proposed method achieves better results in classifying small molecule compounds by mass spectrum data. The accuracy of the test set is increased to $98.894 \%$, and the loss function value is reduced to 0.0502. This model effectively solves the problem of underfitting when using the DNN and the fusion models of CNN and DNN for data classification and improves the prediction accuracy and search efficiency for the mass spectrum database.