Graph Topology Reframes the Coherence of Cell-State Manifold Inference under Heterogeneous Single-Cell Observations.

Area of Science:

• Computational biology
• Single-cell omics analysis
• Bioinformatics

Background:

• Manifold-based analyses in single-cell omics assume low-dimensional representations of biological constraints.
• Real-world single-cell data exhibit heterogeneity, with coexisting shallowly and deeply observed cells.

Purpose of the Study:

• To investigate the impact of observation heterogeneity on manifold-based inference in single-cell omics.
• To develop methods for reliable manifold inference despite data heterogeneity.

Main Methods:

• Analysis of empirical single-cell RNA sequencing datasets.
• Graph abstraction techniques applied to both heterogeneous and homogeneous cell subsets.
• Simulations to model the effects of heterogeneous observations.
• Proposal of topological stability descriptors for manifold skeletons.

Main Results:

• Shallowly observed cells in empirical data form spurious hubs, creating illusory loops in manifold skeletons.
• Imputation methods often fail to correct these artifacts.
• Graph abstraction on homogeneous cells recovers tree-like structures representing cell state transitions.
• Simulations confirm that heterogeneity generates spurious subclusters and false branching.

Conclusions:

• Observation heterogeneity is a systemic distortion, not just noise, in manifold-based inference.
• Topological stability descriptors can identify trustworthy regimes for inference.
• Addressing observation heterogeneity is crucial for accurate single-cell omics analysis.