Interpretable single-cell factor decomposition using sciRED.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) reveals cellular heterogeneity but faces challenges from technical noise, sparsity, and high dimensionality.
• Existing data factorization methods require manual interpretation of identified gene expression programs.
• Improved interpretability is crucial for extracting meaningful biological insights from scRNA-seq data.

Purpose of the Study:

• To develop a novel method, Single-Cell Interpretable Residual Decomposition (sciRED), for enhanced interpretation of scRNA-seq factor analysis.
• To improve the identification and characterization of biological signals within complex scRNA-seq datasets.
• To address confounding technical factors and uncover hidden biological phenomena.

Main Methods:

• sciRED employs residual decomposition to remove confounding effects and uses factor rotations for improved interpretability.
• The method maps identified factors to known covariates and detects unexplained factors potentially representing novel biological signals.
• Genes and biological processes associated with resulting factors are systematically determined.

Main Results:

• Application of sciRED to diverse scRNA-seq datasets revealed sex-specific variation in kidney data.
• The method discerned varying immune stimulation signals in peripheral blood mononuclear cell (PBMC) data.
• sciRED reduced ambient RNA contamination in rat liver data and identified rare cell type signatures and zonation programs in human liver data.

Conclusions:

• sciRED significantly enhances the interpretability of scRNA-seq factor analysis.
• The method effectively characterizes diverse biological signals, including technical variations and subtle biological phenomena.
• sciRED offers a valuable tool for deeper biological discovery in scRNA-seq studies.