Con-AAE

Con-AAE aligns and integrates single-cell RNA sequencing (scRNA-seq) and single-cell Assay for Transposase-Accessible Chromatin using sequencing (scATAC-seq) datasets by mapping them into a coordinated latent subspace to enable multi-omics analysis.

Key Features:

• Adversarial autoencoder architecture: Uses adversarial autoencoder techniques to learn mappings between modalities.
• Contrastive learning: Employs contrastive learning principles to encourage discriminative representations.
• Cycle-consistency constraints: Applies cycle-consistency constraints within the adversarial framework to preserve cross-modal correspondences.
• Latent coordinated subspace mapping: Maps high-dimensional scRNA-seq and scATAC-seq data into a coordinated latent subspace for alignment.
• Noise and sparsity robustness: Addresses high sparsity, high dimensionality, and experimental noise in single-cell data.
• PyTorch implementation: Implemented in PyTorch.
• Discriminative and robust representations: Produces representations that are both discriminative and robust to noise.
• Benchmark evaluation: Demonstrated performance on multiple benchmark datasets.

Scientific Applications:

• Single-cell multi-omics integration: Integrates scRNA-seq and scATAC-seq to enable combined analyses of transcriptomic and chromatin accessibility data.
• Cross-modality alignment: Aligns corresponding cellular states across omic modalities for comparative analyses.
• Dimensionality reduction for noisy data: Reduces dimensionality of sparse, high-dimensional single-cell datasets while preserving biological signal.
• Latent representation for interpretation: Produces coordinated latent representations that facilitate interpretation of multi-omics datasets.

Methodology:

Implements adversarial autoencoders combined with contrastive learning and cycle-consistency constraints to map scRNA-seq and scATAC-seq into a coordinated latent subspace; implemented in PyTorch.