iMAP-dl

iMAP-dl integrates single-cell RNA-sequencing datasets using adversarial paired transfer networks, deep autoencoders, and generative adversarial networks (GANs) to remove batch effects, detect batch-specific cells, and harmonize shared cell-type distributions across platforms.

Key Features:

• Unsupervised Batch Effect Removal: An unsupervised framework eliminates batch effects without requiring labeled data.
• Deep Autoencoders and GANs Integration: Combines deep autoencoders with generative adversarial networks to identify batch-specific cells and align shared cell-type distributions.
• Robustness and Scalability: Designed to maintain performance across variations in data quality and scale to large single-cell RNA-seq datasets.

Scientific Applications:

• Integration of heterogeneous single-cell RNA-seq datasets: Harmonizes datasets from multiple sources to enable joint analysis of cellular heterogeneity.
• Tumor microenvironment analysis: Integrates datasets from platforms such as Smart-seq2 and 10x Genomics to reveal cell–cell interactions obscured by batch effects.

Methodology:

iMAP-dl employs an unsupervised framework that uses deep autoencoders and generative adversarial networks within adversarial paired transfer networks to align data distributions from different sources while preserving biological variability, supporting downstream analyses such as clustering, differential expression analysis, and interaction mapping.