scAce

scAce employs a variational autoencoder to jointly learn cell embeddings and cluster assignments from single-cell RNA sequencing (scRNA-seq) gene expression data and adaptively merges clusters to identify cell populations without predefining the number of clusters.

Key Features:

• Variational Autoencoder Integration: Uses a variational autoencoder (VAE) to simultaneously learn low-dimensional cell embeddings and cluster assignments, capturing non-linear structure in scRNA-seq gene expression data.
• Adaptive Cluster Merging: Iteratively merges clusters using an adaptive strategy to refine clustering results without requiring a predefined cluster number.
• Clustering Enhancement: Provides an option to update and refine cluster assignments iteratively by leveraging previous clustering results from other methods.

Scientific Applications:

• Cell type discrimination: Distinguishes between known cell types using learned embeddings and clustering assignments from scRNA-seq data.
• Novel cell-type discovery: Identifies previously uncharacterized cell populations by adaptively refining cluster structure.
• Cellular heterogeneity analysis: Analyzes complex biological systems to resolve heterogeneity at the single-cell level.

Methodology:

Computational steps explicitly include constructing a variational autoencoder to jointly learn embeddings and cluster assignments, applying an adaptive cluster merging procedure to iteratively refine clusters without pre-specifying cluster numbers, and offering an iterative clustering enhancement step that leverages prior clustering results.