Single-cell manifold-preserving feature selection for detecting rare cell populations.

Area of Science:

• Single-cell genomics
• Computational biology
• Molecular biology

Background:

• Detecting rare cell populations (RCPs) and their driving molecular programs is crucial for understanding development, differentiation, and disease.
• Conventional methods struggle to identify defining molecular features of RCPs from high-dimensional, unenriched single-cell data.

Purpose of the Study:

• To develop an unsupervised computational approach for selecting informative molecular features that preserve data structure.
• To enable sensitive detection of both common cell lineages and rare cellular states.

Main Methods:

• Proposed SCMER (Single-Cell Manifold presERving feature selection), an unsupervised algorithm for feature selection.
• Applied SCMER to diverse biological contexts including hematopoiesis, lymphogenesis, tumorigenesis, and drug response.

Main Results:

• SCMER identified a compact set of non-redundant molecular features with clear biological relevance.
• The selected features effectively delineated common cell lineages and rare cellular states across different datasets.
• Demonstrated SCMER's capability in discovering molecular features for high-dimensional single-cell data.

Conclusions:

• SCMER is a powerful tool for discovering defining molecular features of rare cell populations.
• The approach facilitates the design of targeted, cost-effective diagnostic assays and aids in multi-modal data integration.