scAIDE

scAIDE implements an unsupervised deep learning clustering framework to identify and categorize putative and rare cell types and infer developmental trajectories from large-scale single-cell RNA-seq datasets.

Key Features:

• Unsupervised deep learning framework: Integrates an autoencoder-imputation network with a distance-preserved embedding network (AIDE) to learn robust representations from noisy single-cell RNA-seq data.
• Scalability and efficiency: Processes millions of cells and demonstrated analysis of 1.3 million neural cells in 30 minutes.
• Random projection hashing-based k-means: Employs a random projection hashing-based k-means clustering algorithm to enhance detection of rare cell types.
• Identification of cell types and trajectories: Detected 64 clusters mapped to 19 putative cell types and revealed three developmental trajectories among neural stem cells in the neural dataset.
• Application in disease research: Classified subpopulations of malignant cells in a glioblastoma dataset.

Scientific Applications:

• Cellular heterogeneity and development: Characterizing cellular heterogeneity and developmental trajectories across complex tissues using single-cell RNA-seq.
• Rare cell type discovery: Detecting and profiling rare cell types within large cell populations.
• Cancer and translational research: Classifying malignant cell subpopulations in cancers such as glioblastoma to support translational studies and personalized medicine.

Methodology:

Combines an autoencoder-imputation network with a distance-preserved embedding network (AIDE) and applies random projection hashing-based k-means clustering for unsupervised identification of cell clusters from single-cell RNA-seq data.