DANCE

DANCE automates and interprets preprocessing of single-cell data through two modules that discover and recommend optimal pipelines for diverse analytical tasks.

Key Features:

• Method-Aware Preprocessing (MAP): Discovers optimal pipelines for task-specific methods via hierarchical search across normalization, gene selection, and dimensionality reduction steps.
• Dataset-Aware Preprocessing (DAP): Recommends pipelines for new datasets via similarity-based matching to a reference atlas.
• Large-scale Pipeline Search: Executes over 325,000 pipeline searches across six major tasks including clustering, cell type annotation, imputation, joint embedding, spatial domain identification, and cell type deconvolution.
• Interpretable Preprocessing Patterns: Reveals transparent, data-driven preprocessing patterns across tasks, methods, and datasets.

Scientific Applications:

• Single-Cell Clustering: Identifies optimal preprocessing pipelines that improve clustering performance beyond method defaults.
• Cell Type Annotation: Enhances annotation accuracy through task-specific preprocessing recommendations.
• Imputation: Improves missing data recovery by tailoring preprocessing to imputation methods.
• Joint Embedding: Optimizes integration of multiple datasets or modalities via method-aware pipeline selection.
• Spatial Domain Identification: Supports spatial transcriptomics analysis through specialized preprocessing.
• Cell Type Deconvolution: Enhances deconvolution of bulk or spatial data by optimizing preprocessing steps.

Methodology:

DANCE employs hierarchical search in the MAP module and similarity-based matching to a reference atlas in the DAP module to discover and recommend preprocessing pipelines.