scCAN: single-cell clustering using autoencoder and network fusion.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) analysis is crucial for identifying cell types.
• Large cell numbers, high dimensionality, and dropout rates pose significant challenges for scRNA-seq data analysis.
• Accurate cell type identification requires robust clustering methods.

Purpose of the Study:

• To introduce a novel computational method, scCAN, for unsupervised clustering of scRNA-seq data.
• To address the challenges of scalability, high dimensionality, and sparsity in scRNA-seq data.
• To provide an accurate and efficient tool for cell type segregation.

Main Methods:

• Developed single-cell Clustering using Autoencoder and Network fusion (scCAN).
• Utilized autoencoder and network fusion for data integration and dimensionality reduction.
• Validated scCAN on 28 real scRNA-seq datasets (>3 million cells) and 243 simulated datasets.
• Compared scCAN performance against state-of-the-art methods: CIDR, SEURAT3, Monocle3, SHARP, and SCANPY.

Main Results:

• scCAN accurately estimates the number of true cell types.
• The method effectively segregates cells into distinct types.
• scCAN demonstrates robustness against dropout events, a common issue in scRNA-seq data.
• scCAN exhibits superior speed and memory efficiency compared to existing methods.
• Extensive validation shows scCAN outperforms other leading scRNA-seq clustering tools in accuracy and scalability.

Conclusions:

• scCAN is a highly accurate and scalable method for unsupervised clustering of large-scale scRNA-seq data.
• The developed method effectively overcomes key challenges in single-cell data analysis, including sparsity and high dimensionality.
• scCAN provides a valuable tool for cell type identification and biological discovery from scRNA-seq experiments.