Research on intelligent evaluation algorithm of genetic variation pathogenicity for multiomics data fusion

In this study, an intelligent evaluation algorithm of genetic variation pathogenicity oriented to multi-omics data fusion-Dynamic Multi-omics Fusion Network (DMF-NET) is proposed. Through dynamic weight fusion strategy, the algorithm adaptively adjusts the weights of genome, transcriptome and epigenome data according to different mutation types. At the same time, it introduces multi-headed cross-genomics attention mechanism to mine the complex interaction between genomics, and combines SHAP value analysis and biological knowledge base to enhance the clinical interpretability of the model. The experiment uses several mainstream monocomponent models as the baseline comparison. The results show that DMF-Net is significantly superior to the existing methods in terms of AUROC, AUPRC, sensitivity, specificity and F1 score, indicating that DMF-NET has higher accuracy and adaptability in pathogenicity assessment. In addition, the dynamic weight analysis shows that the model can flexibly adjust the omics weight according to different mutation types, and the cross-omics attention analysis reveals the potential pathogenic regulation path. The high consistency between the results of SHAP analysis and the clinical knowledge base further verifies the clinical interpretability of the model. This study provides an efficient and interpretable new method for the pathogenicity evaluation of genetic variation, which is expected to promote the development of precision medicine.