Protein group data analysis and disease prediction based on convolutional neural network and transformer model

In this study, a data analysis framework of protein group combined with Convolutional Neural Network (CNN) and Transformer model is proposed to improve the analytical ability of protein group data and the accuracy of disease prediction. Mass spectrometry data were collected and preprocessed from CPTAC and PRIDE Archive databases, and the framework of Hybrid CNN-Transformer was adopted, integrating technological innovations such as multi-scale feature extraction, sequence-structure joint modeling and dynamic feature fusion. AdamW optimizer, cosine annealing learning rate scheduling and other training strategies are used, and Focal Loss and Dice Loss are combined to optimize the loss function. The experimental results show that the CNN-Transformer fusion model is superior to the pure CNN and Transformer architecture in terms of ROC-AUC, F1-score and Matthews correlation coefficient (MCC) on the test set, while maintaining a reasonable reasoning speed. The ablation experiment further verified the important contribution of each component of the model to the performance. The spatial distribution of t-SNE features shows the clear separation of health and disease samples, and the cross-modal attentional thermogram reveals the strong correlation between the model and key biomarker areas. This study provides an efficient and accurate new method for protein group data analysis and disease prediction, and shows the obvious advantages of the hybrid model in dealing with complex biological information.