Condition-Associated Pattern Extraction and Recovery From Multi-Condition Single-Cell RNA-seq Data With CAPER.

Area of Science:

• Genomics
• Computational Biology
• Immunology

Background:

• Analyzing multi-condition single-cell RNA sequencing (scRNA-seq) data is challenging due to the difficulty in distinguishing true biological signals from experimental variations.
• Existing methods often fail to resolve cell-type-specific responses or may over-correct, removing valuable biological information.

Purpose of the Study:

• To introduce CAPER, a novel matrix factorization framework designed to disentangle shared biological states from condition-specific variations in scRNA-seq data.
• To develop a tool that outputs an interpretable, batch-corrected expression matrix, preserving and isolating biological signals of interest.

Main Methods:

• CAPER utilizes a matrix factorization approach to explicitly separate biological signals from technical noise.
• The framework was validated through extensive simulations and applied to three real-world scRNA-seq datasets with varying signal-to-noise ratios (SNRs).

Main Results:

• CAPER successfully generated interpretable latent factors associated with biological relevance across different SNR scenarios.
• The method accurately identified key differentially expressed genes and responsive cell populations in immune stimulation, tumor microenvironment, and autoimmune disease datasets.
• CAPER provided a robust, batch-corrected expression matrix, isolating true biological signals.

Conclusions:

• CAPER is a robust and interpretable tool for analyzing multi-condition scRNA-seq data.
• It enables reliable recovery of biological signals, facilitating discoveries in disease research and functional genomics by accurately disentangling biological variation from technical artifacts.