scANMF: Prior Knowledge and Graph-Regularized NMF for Accurate Cell Type Annotation in scRNA-seq.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) offers high-resolution cellular heterogeneity insights.
• Accurate cell-type annotation is hindered by data sparsity, noise, and cross-platform/species variability.
• Existing tools often rely on single prior knowledge sources, limiting performance with incomplete data.

Purpose of the Study:

• To develop a robust framework for accurate cell-type annotation in scRNA-seq data.
• To integrate diverse prior knowledge sources for improved annotation performance.
• To address limitations of existing annotation tools, especially with restricted prior information.

Main Methods:

• Introduced scANMF, a prior- and graph-regularized non-negative matrix factorization framework.
• Integrated marker-gene information, partial label supervision, and local manifold structure.
• Factorized expression matrices into interpretable gene-factor and cell-factor representations.

Main Results:

• Achieved high annotation accuracy across within-dataset, cross-platform, and cross-species evaluations.
• Demonstrated stability under varying label sparsity and marker-gene noise.
• Showcased robustness to hyperparameter choices and complementary contributions from different priors.

Conclusions:

• scANMF provides a practical and robust framework for scRNA-seq data annotation.
• The method excels particularly when high-quality prior knowledge is limited.
• Integration of multiple priors significantly enhances annotation accuracy and stability.