SCALE

SCALE extracts interpretable latent features from single-cell ATAC-seq (scATAC-seq) chromatin accessibility data to enable denoising, clustering, visualization, and identification of batch effects.

Key Features:

• Deep Generative Framework: Employs a deep generative framework to model high-dimensional, sparse scATAC-seq signals and extract latent features representing chromatin accessibility.
• Probabilistic Gaussian Mixture Model (GMM): Integrates a probabilistic Gaussian mixture model to characterize latent-feature distributions and support clustering and denoising.
• Interpretable Features: Produces interpretable latent features that correlate with specific cell populations and chromatin accessibility patterns.
• Batch Effect Identification: Detects and highlights batch effects across scATAC-seq experiments.

Scientific Applications:

• Visualization: Provides low-dimensional embeddings for visualizing chromatin accessibility landscapes from scATAC-seq data.
• Clustering: Enables identification of distinct cell populations via latent-space clustering of scATAC-seq profiles.
• Denoising and Imputation: Performs denoising and imputes missing accessibility signals to improve downstream analyses of scATAC-seq datasets.

Methodology:

Implemented in PyTorch and based on a deep generative framework combined with a probabilistic Gaussian mixture model for latent feature learning.