CIRCE

CIRCE predicts cis-regulatory DNA interactions from single-cell chromatin accessibility data by inferring co-accessibility networks.

Key Features:

• Scalable implementation: Re-implements the Cicero workflow with runtime and memory use reduced by several orders of magnitude for single-cell atlas analysis.
• Metacell computation: Groups similar cells to reduce data sparsity, with demonstrated improvement using promoter capture Hi-C data.
• Flexible pre-processing: Evaluates impact of different pre-processing approaches on DNA interaction predictions, with best performance obtained using the single-cell count matrix directly without Cicero's count normalization.
• Large-scale processing: Processes datasets of more than 700000 cells and 1 million DNA regions in less than an hour.

Scientific Applications:

• Cis-regulatory interaction mapping: Predicts DNA region interactions from single-cell ATAC-seq data for gene expression regulation studies.
• Single-cell atlas analysis: Enables co-accessibility network inference on large-scale single-cell chromatin accessibility datasets.
• Benchmarking and validation: Demonstrated against Cicero on datasets of different sizes and validated with promoter capture Hi-C data.

Methodology:

CIRCE utilizes skggm for graphical lasso and scanpy for data processing. The workflow re-implements the Cicero approach with optional metacell computation to reduce data sparsity.