Milo

Milo performs differential abundance testing on single-cell datasets using k-nearest neighbor (k-NN) graphs to detect changes in cell-state abundance across continuous cellular manifolds such as differentiation trajectories.

Key Features:

• Flexible statistical framework: Implements a flexible and scalable statistical framework that assigns cells to partially overlapping neighborhoods within a k-NN graph.
• Neighborhood sampling and refinement: Samples and refines neighborhoods across the k-NN graph to capture subtle biological variation.
• Generalized linear models (GLMs): Uses GLMs to perform differential abundance testing across varied experimental settings.
• Manifold-aware analysis: Operates on continuous cellular manifolds rather than relying on predefined discrete clusters.
• Robustness and sensitivity: Demonstrates robustness and sensitivity in simulations for detecting cell-state perturbations.

Scientific Applications:

• Disease and developmental abundance shifts: Detects shifts in the abundance of cell states associated with disease conditions or developmental processes without predefined clusters.
• Aging and thymic epithelial precursor differentiation: Identified perturbed differentiation during aging in a lineage-biased thymic epithelial precursor state.
• Human cirrhotic liver perturbations: Uncovered extensive perturbations across multiple lineages in human cirrhotic liver.

Methodology:

Constructs k-nearest neighbor (k-NN) graphs, assigns cells to partially overlapping neighborhoods, samples and refines neighborhoods across the graph, applies generalized linear models (GLMs) for differential abundance testing, and uses simulations for evaluation.