A knowledge-guided approach to recovering important rare signals from high-dimensional single-cell data

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell transcriptomic data is high-dimensional, posing challenges for analysis.
• Current dimensionality reduction methods may overlook rare but significant biological signals.
• Identifying rare cell subtypes and distinguishing similar cell populations is crucial in biological research.

Purpose of the Study:

• To develop a novel framework for dimensionality reduction in single-cell transcriptomic data.
• To guide dimensionality reduction using knowledge-derived genes of interest.
• To improve the identification and clustering of rare cells and highly similar cell sub-populations.

Main Methods:

• Proposed a novel framework for dimensionality reduction.
• Integrated knowledge-derived genes of interest to guide the reduction process.
• Applied the framework to analyze single-cell transcriptomic datasets.

Main Results:

• The framework successfully identified endocrine cell subtypes in the pancreatic islet.
• Demonstrated the ability to separate highly similar hematopoietic sub-populations.
• Enabled the detection of rare senescent cells.

Conclusions:

• The proposed framework enhances the interpretability of high-dimensional single-cell data.
• Guiding dimensionality reduction with specific genes of interest is effective for discovering rare and subtle biological signals.
• This approach offers improved capabilities for cell sub-population analysis in various biological contexts.