Systematic background selection with BasCoD enhances contrastive dimension reduction in single cell genomics.

Area of Science:

• Single-cell genomics
• Computational biology
• Statistical inference

Background:

• Distinguishing condition-specific variation from shared variation is vital in single-cell studies.
• Ultra-high-dimensional single-cell data requires effective dimension reduction for biological insights.
• Contrastive dimension reduction methods rely heavily on appropriate background dataset selection.

Purpose of the Study:

• To introduce BasCoD, a novel statistical framework for evaluating and selecting background datasets in contrastive dimension reduction.
• To address the lack of formal criteria for background selection in single-cell data analysis.
• To improve the contrast and interpretability of single-cell data representations.

Main Methods:

• Developed BasCoD based on spectral subspace inclusion theory.
• Applied BasCoD to diverse single-cell datasets.
• Utilized BasCoD to guide contrastive analysis design and elucidate interaction effects.

Main Results:

• BasCoD effectively identifies suitable background datasets.
• The use of BasCoD-selected backgrounds substantially improves the contrast and interpretability of target representations.
• BasCoD facilitates the design of large-scale single-cell experiments under heterogeneous conditions.

Conclusions:

• BasCoD provides a rigorous statistical approach for background selection in contrastive dimension reduction.
• This framework enhances the biological interpretation of treatment-specific responses in single-cell data.
• BasCoD is valuable for designing complex single-cell experiments and analyzing perturbation studies.