Differential abundance testing on single-cell data using k-nearest neighbor graphs.

Area of Science:

• Computational biology
• Single-cell genomics
• Statistical modeling

Background:

• Current single-cell analysis methods often rely on discrete clusters, which may lack the resolution to detect subtle biological changes or continuous cellular processes.
• This clustering approach can obscure important biological perturbations and hinder accurate differential abundance testing.

Purpose of the Study:

• To introduce Milo, a novel statistical framework for scalable differential abundance testing in single-cell datasets.
• To overcome the limitations of cluster-based analyses by utilizing cell-cell similarity structures.
• To provide a robust method for identifying biological perturbations in single-cell experiments.

Main Methods:

• Milo employs a k-nearest neighbor graph to define partially overlapping cell neighborhoods for analysis.
• The framework performs differential abundance testing on these neighborhoods, allowing for finer resolution than discrete clusters.
• Validated using simulations and single-cell RNA sequencing (scRNA-seq) data from various biological contexts.

Main Results:

• Milo successfully identifies biological perturbations that are missed by traditional cluster-based methods.
• The framework demonstrates robust control of the false discovery rate, even in the presence of batch effects.
• Milo outperforms existing differential abundance testing strategies in detecting biologically relevant changes.

Conclusions:

• Milo offers a more sensitive and accurate approach to differential abundance testing in single-cell genomics.
• The method is applicable to various single-cell data types beyond scRNA-seq due to its reliance on cell-cell similarity.
• Milo is available as an open-source R package, facilitating its adoption in the research community.