Hierarchicell

Hierarchicell estimates statistical power for differential expression tests in single-cell RNA-seq by simulating hierarchical correlation structures and is implemented as an R package.

Key Features:

• Hierarchical correlation structure: Explicitly simulates and accounts for within-sample (within-individual) correlations so that individuals or samples are treated as the experimental units rather than aggregated cell counts.
• Simulation-based modeling: Employs a multi-stage simulation process to model gene dropout rates, intra-individual dispersion, and inter-individual variation.
• Variable or fixed cell counts per individual: Handles scenarios with either variable or fixed numbers of cells per individual in simulations.
• Correlation modeling among cells: Models correlation among cells within an individual to provide realistic representation of single-cell data structure.
• Phenotype support: Supports power estimation for both binary and continuous phenotypes.
• Configurable experimental units and cell counts: Allows specification of the number of independent experimental units and the number of cells per unit.

Scientific Applications:

• Experiment design: Estimating required numbers of independent samples and cells per sample to achieve target power for differential expression in scRNA-seq studies.
• Robustness and reproducibility assessment: Evaluating how hierarchical correlation and sample-level variability affect false-positive rates and power estimates.
• Discovery potential evaluation: Assessing the likelihood of detecting biologically meaningful differential expression signals under specified experimental designs and dropout/dispersion scenarios.

Methodology:

Performs multi-stage simulations that simulate gene dropout rates, model intra-individual dispersion and inter-individual variation, allow variable or fixed numbers of cells per individual, and model correlations among cells within individuals to generate power estimates for differential expression tests.