CarDEC

CarDEC performs simultaneous clustering, denoising, and batch-effect correction of single-cell RNA sequencing (scRNA-seq) data in both embedding and gene expression spaces.

Key Features:

• Algorithm: Implements a count-adapted regularized Deep Embedded Clustering framework using deep learning for scRNA-seq data.
• Joint clustering and denoising: Simultaneously clusters cells and denoises expression profiles to increase signal quality for downstream analysis.
• Batch correction scope: Corrects batch effects in both low-dimensional embedding spaces and the original gene expression space.
• Multi-batch scalability: Efficiently handles multiple batches and avoids the two-batch limitation of Seurat 3.0's mutual nearest neighbor (MNN) method.
• Non-HVG information recovery: Denoising increases information content so non-highly variable genes (non-HVGs) can contribute comparably to clustering as highly variable genes (HVGs).
• Trajectory and regulator discovery: Produces batch-corrected, denoised expression profiles that facilitate trajectory analysis and identification of marker genes and transcription factors.
• Benchmarking: Demonstrated superior performance relative to Scanorama, DCA + Combat, scVI, and MNN across various species and tissues.
• Computational efficiency: Optimized for large-scale scRNA-seq datasets to reduce computational burden.

Scientific Applications:

• Cell-type clustering: Improved clustering of single cells from scRNA-seq experiments via integrated denoising and embedding.
• Batch-effect correction: Removal of systematic batch differences in both embedding and gene expression spaces for more accurate biological interpretation.
• Denoising and gene usage expansion: Enabling inclusion of non-HVGs in analyses to provide a more comprehensive view of cellular heterogeneity.
• Trajectory and regulatory analysis: Facilitating reconstruction of developmental or differentiation trajectories and discovery of marker genes and transcription factors.
• Cross-species and tissue benchmarking: Comparative evaluation of methods and datasets across various species and tissues.
• Large-scale scRNA-seq studies: Scalable analysis for high-throughput single-cell datasets.

Methodology:

Uses a count-adapted, regularized Deep Embedded Clustering (deep learning) approach that simultaneously performs clustering and denoising while correcting batch effects in both embedding and gene expression spaces.