Augur

Augur prioritizes cell types that are most responsive to experimental perturbations in high-dimensional single-cell molecular datasets.

Key Features:

• Machine learning framework: Trains classifiers to distinguish cells from treated versus control conditions using molecular measurements.
• Cell type–specific classifiers: Builds and evaluates a separate classifier for each annotated cell type to measure its responsiveness to perturbations.
• Quantitative separability assessment: Uses predictive accuracy as a quantitative metric of separability between perturbed and unperturbed cells.
• Cross-validation: Employs cross-validation to evaluate classifier performance and provide robust estimates of predictive accuracy.
• Supported molecular modalities: Applied to single-cell RNA sequencing and chromatin accessibility datasets (and compatible single-cell molecular assays).

Scientific Applications:

• Single-cell RNA sequencing: Identifies cell types with significant changes in gene expression profiles following experimental interventions.
• Chromatin accessibility: Detects cell types showing regulatory changes in chromatin accessibility in response to perturbations.
• Imaging transcriptomics: Applies to imaging-based transcriptomic datasets to quantify spatially resolved cellular responses.
• Method comparison: Demonstrates performance relative to differential gene expression methods for prioritizing responsive cell types.
• Neural circuit analysis in mice: Has been used to identify neural circuits implicated in restoring locomotion in mice following spinal cord neurostimulation.

Methodology:

Trains cell type–specific classifiers on molecular measurements (e.g., single-cell RNA sequencing and chromatin accessibility) to predict experimental sample labels (treatment vs control) and quantifies separability via cross-validation of predictive accuracy.