Ensemble deep learning of embeddings for clustering multimodal single-cell omics data.

Area of Science:

• Single-cell omics
• Computational biology
• Bioinformatics

Background:

• Multimodal single-cell omics technologies profile multiple molecular attributes (gene expression, chromatin accessibility, protein abundance) simultaneously.
• Extracting integrated information across modalities for accurate cell clustering remains a computational challenge.

Purpose of the Study:

• To develop an unsupervised ensemble deep learning framework, SnapCCESS, for effective integration and clustering of multimodal single-cell omics data.
• To improve cell clustering and characterization by leveraging information across multiple molecular data modalities.

Main Methods:

• SnapCCESS utilizes variational autoencoders to create embeddings of multimodal data.
• It employs an unsupervised ensemble deep learning framework for consensus clustering.
• The method is compatible with various clustering algorithms.

Main Results:

• SnapCCESS effectively integrates multimodal single-cell omics data for cell clustering.
• It demonstrates superior efficiency and performance compared to conventional ensemble deep learning methods.
• The framework outperforms existing multimodal embedding generation methods in data integration for clustering.

Conclusions:

• SnapCCESS provides a robust and efficient solution for multimodal single-cell omics data integration and clustering.
• Improved cell clustering facilitates more accurate cell identity and type characterization.
• This advancement is crucial for downstream analyses in single-cell omics research.