distinct

distinct performs differential testing between two or more groups of distributions via hierarchical non-parametric permutation tests on cumulative distribution functions to identify both mean-based and non-mean-based differential patterns in single-cell and other high-throughput data.

Key Features:

• Hierarchical non-parametric permutation testing: Differential testing is performed via hierarchical non-parametric permutation tests on the cumulative distribution functions of each sample.
• Full distribution comparison: Comparing empirical cumulative distribution functions identifies differential patterns involving changes in the mean as well as more subtle variations that do not involve the mean, such as unimodal versus bi-modal distributions with the same mean.
• Fully non-parametric framework: The method makes no assumptions on how the data was generated, allowing application to a variety of datasets.
• Multiple group support: The method can perform differential testing between two or more groups of samples, with at least two samples (biological replicates) per group required.

Scientific Applications:

• Single-cell RNA sequencing differential analysis: distinct identifies complex variations between conditions in scRNA-seq data where mean-based methods struggle, including changes where the mean is only marginally affected.
• High-dimensional flow and mass cytometry analysis: The method is well suited for differential state analyses within sub-populations of cells in HDCyto data.
• Differential state analysis: distinct performs differential analyses within sub-populations of cells to identify condition-specific cellular states.

Methodology:

distinct employs a hierarchical nonparametric permutation approach comparing empirical cumulative distribution functions between groups of samples. Extensive benchmarks were performed across both simulated and experimental datasets from single-cell RNA sequencing and mass cytometry data, showing favorable performance, identification of more differential patterns than competitors, and good control of false positive and false discovery rates.