scIAE

scIAE provides an integrative autoencoder-based ensemble classification framework to extract robust low-dimensional representations from single-cell RNA sequencing (scRNA-seq) data for applications such as cell type annotation and disease status prediction.

Key Features:

• Multiple random projections: Applies multiple random projections to reduce dimensionality while preserving essential features of scRNA-seq gene expression profiles.
• Autoencoder integration: Integrates stacked, denoising, and sparse autoencoders to generate compressed representations of gene expression data.
• Ensemble classification: Uses base classifiers trained on the lower-dimensional representations and integrates their predictions within an ensemble framework.
• Dropout and sparsity handling: Employs denoising and sparse autoencoders to mitigate effects of sparsity and dropout events in scRNA-seq data.
• Dimensionality-robust performance: Delivers robust feature extraction performance across a range of chosen dimensionalities and reportedly outperforms common feature extraction methods.

Scientific Applications:

• Cell type annotation: Generates representations suitable for annotating cell types within scRNA-seq datasets.
• Cross-batch comparisons: Facilitates comparisons across batches by producing stable low-dimensional embeddings.
• Platform-independent analyses: Supports analyses that are resilient to differences between sequencing platforms.
• Interspecies studies: Enables comparative analyses across species using compressed gene expression representations.
• Disease status prediction: Provides representations and ensemble classification for predicting disease status from scRNA-seq data.
• Cell atlas construction: Supplies compact and informative features useful for building comprehensive cell atlases.

Methodology:

Performs multiple random projections for dimensionality reduction, integrates stacked, denoising, and sparse autoencoders to produce compressed representations, trains base classifiers on those representations, and integrates predictions from multiple models within an ensemble.