Application of multi-omics fusion technique based on Raman spectroscopy and metabolomics in early diagnosis and activity prediction of systemic lupus erythematosus

The combination of artificial intelligence and Raman spectroscopy provides new ideas and methods for auxiliary diagnosis of diseases. However, in systemic lupus erythematosus (SLE), there are problems of high pathological consistency and large overlap of spectral information, and single spectral omics cannot obtain ideal results. However, metabolomics has the advantages of directly reflecting the metabolic status in organisms and gaining in-depth understanding of the physiological and pathological states of organisms. At the same time, multi-omics fusion technology can effectively integrate the characteristics of different omics levels. Therefore, this study proposed a Multi-omics Decoupling-Bipartite Attentional Weighting (MDBAW) fusion model based on Raman spectroscopic omics and metabolomics data for the first time. The model fully considers the unique and shared representations between omics, and adds attention weight modules at the input and output ends to give more weight to the features with large amount of information in the two omics modalities. Finally, the experimental results on three data sets proved that the MDBAW model is superior to single-omics and other advanced multi-omics fusion models, and can effectively improve the accuracy of SLE classification diagnosis and activity prediction. In addition, through the correlation analysis of Raman spectroscopic omics and metabolomics data and KEGG pathway analysis, the interpretability of the fusion of these two omics in auxiliary disease diagnosis applications was verified, and the ability of Raman spectroscopy to detect metabolites was proved.