scMSAC Assigns Single-Cell Multi-Omics Data at the Multi-Modal Cluster via Subgraph Attention Autoencoder.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell multi-omics sequencing allows simultaneous measurement of multiple data types from individual cells.
• Joint clustering of this data is crucial for understanding cell states and molecular mechanisms in oncology, neurology, and developmental biology.
• Challenges include feature space disparities and data noise, hindering accurate clustering.

Purpose of the Study:

• To introduce scMSAC, a novel clustering method for single-cell multi-omics data.
• To address challenges posed by feature space differences and data noise in multi-omics data integration.
• To improve the accuracy and comprehensiveness of cell state depiction and molecular mechanism discovery.

Main Methods:

• scMSAC utilizes a denoising subgraph attention autoencoder for clustering single-cell multi-omics data.
• A weighted nearest neighbor graph strategy assigns weights to multi-omics data, creating a comprehensive similarity graph.
• Spatial Channel Attention (SCA) mechanism fuses omics features, reducing discrepancies and enhancing clustering.

Main Results:

• scMSAC demonstrates excellent clustering performance compared to existing methods.
• The method excels in identifying rare cell types.
• scMSAC performs effectively in differential expression analysis.

Conclusions:

• scMSAC offers a robust approach for single-cell multi-omics data clustering.
• The method successfully integrates diverse omics data, overcoming feature space challenges.
• scMSAC provides significant advancements for biological research, particularly in oncology and neurology.