Deep learning enables accurate clustering with batch effect removal in single-cell RNA-seq analysis.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) enables cell type and state characterization via unsupervised clustering.
• Increasing cell numbers and batch effects present significant computational challenges in scRNA-seq data analysis.

Purpose of the Study:

• To introduce DESC, an unsupervised deep embedding algorithm for clustering scRNA-seq data.
• To address computational challenges in scRNA-seq analysis, including batch effect removal and scalability.

Main Methods:

• DESC employs iterative optimization of a clustering objective function for unsupervised learning.
• The algorithm performs iterative self-learning to gradually remove batch effects.
• DESC functions as a soft clustering algorithm, providing interpretable cluster assignment probabilities.

Main Results:

• DESC effectively clusters scRNA-seq data while removing batch effects.
• The algorithm demonstrates a balance between clustering accuracy and stability.
• DESC has a low memory footprint and can leverage GPU acceleration.
• Batch information is not explicitly required for DESC's batch effect removal.

Conclusions:

• DESC offers a valuable tool for analyzing large-scale scRNA-seq datasets.
• The algorithm aids biomedical researchers in understanding cellular heterogeneity.
• DESC's soft clustering provides insights into discrete and pseudotemporal cell structures.