An improved multiomics data clustering algorithm based on graph regularization constraints

Accurate clustering of cell type classification by integrating multi-omics data is necessary for subsequent data analysis. However, related algorithms face difficulties in dealing with noise and missing values present in multi-omics data. The network-based integrative clustering algorithm (NIC) can overcome the problem of data heterogeneity but does not fully utilize the spatial structure information in the data. To solve this problem, this study proposes a new algorithm-graph regularization-based NIC (GR-NIC) to improve the integration accuracy, remove redundant features, and preserve the geometric structure of the data by fusing single-cell multi-omics data, single-cell sequencing data (scRNA-seq or scATACseq) and histone modification data. To verify the effectiveness of the GR-NIC algorithm, this paper compares three state-of-the-art multi-omics clustering algorithms through five datasets. The results show that GR-NIC can select features that are more representative of the structure of the data and improve the accuracy of clustering. This new approach is effectively validated on mouse brain and peripheral blood sample datasets and shows significant improvements.