Clustering single-cell multi-omics data with MoClust.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Single-cell multi-omics sequencing generates high-dimensional, sparse data with potential doublets.
• Diverse data distributions across omics hinder accurate clustering and cellular heterogeneity dissection.
• Existing methods struggle with data alignment and quality control for multi-omics datasets.

Purpose of the Study:

• To develop a novel joint clustering framework, MoClust, for single-cell multi-omics data analysis.
• To address challenges of high dimensionality, sparsity, diverse distributions, and doublets in multi-omics data.
• To improve the accuracy and interpretability of clustering for dissecting cellular heterogeneity.

Main Methods:

• Developed MoClust, a joint clustering framework incorporating omics-specific autoencoders.
• Integrated a selective automatic doublet detection module for data quality improvement.
• Employed contrastive learning for distribution alignment, creating an omics-invariant representation.

Main Results:

• MoClust demonstrated powerful alignment, doublet detection, and clustering capabilities on simulated and real datasets.
• The distribution alignment approach enhanced cluster compactness and separability.
• The framework adaptively weighted omics contributions for robust clustering.

Conclusions:

• MoClust offers a robust solution for single-cell multi-omics clustering, improving cellular heterogeneity analysis.
• The framework effectively handles data complexities including doublets and diverse omics distributions.
• MoClust provides a valuable tool for researchers in computational biology and genomics.