scAlign

scAlign aligns and integrates single-cell RNA sequencing (scRNA-seq) datasets using an unsupervised deep learning approach to estimate per-cell differences in gene expression across conditions, tissues, or species.

Key Features:

• Unsupervised deep learning: Employs an unsupervised neural-network framework to learn alignments between scRNA-seq datasets without requiring full prior cell-type labels.
• Per-cell difference estimation: Estimates per-cell differences in gene expression to quantify cellular heterogeneity across datasets.
• Incorporation of cell labels: Can incorporate partial, overlapping, or complete sets of cell labels to guide alignment when available.
• Batch effect correction: Aligns datasets to mitigate batch effects so that expression differences reflect biological variation rather than technical artifacts.
• Rare cell identification: Detects gene expression programs within rare cell populations, demonstrated in applications such as malaria parasites.

Scientific Applications:

• Integration of scRNA-seq datasets: Merges datasets from different conditions, tissues, or species to enable comparative analyses.
• Batch effect correction: Removes technical variation across experiments to reveal biologically relevant expression patterns.
• Identification of cell type-specific expression differences: Uncovers subtle gene expression differences across cell types within and between datasets.
• Analysis of rare populations: Characterizes gene expression programs in rare cell populations, including examples in malaria parasite studies.

Methodology:

Uses an unsupervised deep learning framework to align and integrate scRNA-seq data, estimate per-cell gene expression differences, and optionally incorporate partial/overlapping/complete cell labels to accommodate varying dataset composition.