Sensitive cluster-free differential expression testing.

Area of Science:

• Genomics
• Computational Biology
• Single-cell analysis

Background:

• Differential gene expression (DE) analysis is crucial for characterizing disease phenotypes using single-cell RNA sequencing (scRNA-seq) data.
• Current DE pipelines rely on discrete cell clustering, which limits sensitivity and specificity, especially for continuous biological processes.
• The granularity of cell type annotation in clustering-based approaches is a key limitation for accurate DE testing.

Approach:

• We introduce miloDE, a cluster-free framework for DE testing in scRNA-seq data.
• miloDE adapts the Milo approach, utilizing graph-based representations to define overlapping cell neighborhoods.
• The framework incorporates specific adaptations for neighborhood assignment, statistical testing, and multiple testing correction tailored for expression analysis.

Key Points:

• miloDE identifies a transient haemogenic endothelium-like state in Tal1-deficient chimeric mouse embryos.
• The framework reveals distinct transcriptional programs associated with macrophage changes in Idiopathic Pulmonary Fibrosis.
• Demonstrated performance through both simulations and real-world biological datasets.

Conclusions:

• miloDE offers a more sensitive and specific approach to DE analysis compared to cluster-based methods.
• The cluster-free framework is particularly advantageous for analyzing continuous biological trajectories.
• miloDE provides an open-source R package, facilitating broader adoption in scRNA-seq research.