A multiresolution framework to characterize single-cell state landscapes.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Analyzing cellular heterogeneity in single-cell transcriptomic data presents significant challenges.
• Existing methods struggle to simultaneously identify cell states and their topological structures effectively.

Purpose of the Study:

• To introduce multiresolution cell-state decomposition for capturing diverse variability patterns in single-cell data.
• To develop and validate ACTIONet, a framework for robust and interpretable single-cell state characterization.

Main Methods:

• ACTIONet integrates archetypal analysis and manifold learning for multiresolution single-cell state analysis.
• The framework combines dominant pattern discovery with structural representation of the cell state landscape.

Main Results:

• ACTIONet demonstrates superior performance compared to existing methods on synthetic and real datasets.
• The framework successfully integrates and annotates cells across multiple human cortex datasets.
• A consensus vocabulary and gene signatures for human prefrontal cortex cell types and subtypes were defined.

Conclusions:

• ACTIONet provides a powerful, reproducible, and interpretable platform for single-cell transcriptomic analysis.
• The multiresolution approach effectively characterizes cell states and their relationships, advancing biological insights.