Effective multi-modal clustering method via skip aggregation network for parallel scRNA-seq and scATAC-seq data.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Parallel clustering of single-cell RNA-seq (scRNA) and single-cell Assay of Transposase Accessible Chromatin (scATAC) data is increasingly important.
• Existing methods often fail to leverage the richer information in scRNA data compared to scATAC, limiting performance.
• This can compromise the accurate identification of cell subpopulations and tumor microenvironments.

Purpose of the Study:

• To propose an effective multi-modal clustering model, scEMC, for parallel scRNA and scATAC data integration.
• To address the information imbalance between scRNA and scATAC data in clustering.
• To enhance the analysis of cell subpopulations and tumor microenvironments.

Main Methods:

• Developed a skip aggregation network for simultaneous learning of global cell structure and multi-modal integration.
• Implemented a skip connection from scRNA data to safeguard representation quality against sparse scATAC data.
• Utilized a Zero Inflated Negative Binomial-based denoising autoencoder and a joint optimization module with multiple losses.

Main Results:

• The scEMC model demonstrates significant effectiveness in multi-modal clustering tasks.
• Experimental results underscore the model's capability in integrating diverse single-cell data modalities.
• The approach successfully improves the analysis of complex biological systems.

Conclusions:

• scEMC offers an effective solution for parallel clustering of scRNA and scATAC data by addressing modality-specific information content.
• The model advances the accurate identification of cell subpopulations and the characterization of tumor microenvironments.
• The developed methods contribute to the broader field of single-cell multi-omics analysis.