scCAN

scCAN performs unsupervised clustering of single-cell RNA sequencing (scRNA-seq) data to identify cell types and address high dimensionality and dropout in large-scale datasets.

Key Features:

• Autoencoder Integration: scCAN leverages autoencoders to reduce the dimensionality of gene expression matrices while preserving critical biological signals.
• Network Fusion Strategy: scCAN constructs multiple similarity networks from different data aspects and fuses them to capture comprehensive cellular relationships.
• Robustness to Dropout Rates: scCAN demonstrates robustness to high dropout rates in scRNA-seq data, enabling reliable cell-type identification in sparse datasets.
• Scalability and Efficiency: scCAN is designed for fast, memory-efficient processing of large-scale datasets, including analyses involving millions of cells.
• Accurate Cell Type Segregation: scCAN estimates the number of true cell types and accurately segregates cells, validated on 28 real-world scRNA-seq datasets and 243 simulated datasets.

Scientific Applications:

• Cell Type Identification: Identification of putative cell types to characterize cellular diversity within tissues or developmental stages.
• Biomarker Discovery: Facilitation of biomarker discovery associated with specific cell types or states through precise cell segregation.
• Disease Research: Analysis of cellular heterogeneity in disease contexts such as cancer to inform studies of tumor microenvironments.

Methodology:

Data preprocessing including missing-value handling and normalization, dimensionality reduction via autoencoders, construction of multiple similarity networks, network fusion to integrate networks, followed by clustering and validation against known cell types and simulated datasets.