An analytical framework for interpretable and generalizable single-cell data analysis.

Area of Science:

• Computational biology
• Bioinformatics
• Data science

Background:

• Single-cell omics datasets are rapidly growing in diversity and quantity.
• Existing exploratory data analysis methods struggle to scale and generalize across datasets.
• There is a need for interpretable and robust data representations.

Purpose of the Study:

• To develop a scalable and generalizable framework for single-cell omics data analysis.
• To combine the interpretability of linear methods with the representational power of non-linear methods.
• To introduce novel methods for data representation, visualization, and structure discovery.

Main Methods:

• Developed a 'linearly interpretable' framework integrating linear and non-linear modeling.
• Introduced GraphDR for data representation and visualization.
• Introduced StructDR for unified structure discovery (clustering, trajectory, surface estimation).
• Enabled confidence set inference for discovered structures.

Main Results:

• The framework provides interpretable and transferable data representations.
• GraphDR enables effective visualization of complex single-cell data.
• StructDR successfully unifies multiple structure discovery tasks.
• Confidence inference enhances the reliability of structural findings.

Conclusions:

• The developed framework offers a robust and interpretable approach for single-cell omics data analysis.
• GraphDR and StructDR provide powerful tools for exploring and understanding single-cell data heterogeneity.
• This approach facilitates more reliable and generalizable insights from large-scale single-cell datasets.