FaStaNMF: A Fast and Stable Non-Negative Matrix Factorization for Gene Expression

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Gene expression analysis in mixed cell populations yields limited tissue-specific information.
• In silico deconvolution offers a cost-effective alternative to cell sorting or single-cell sequencing for extracting cell type-specific expression data.
• Non-negative matrix factorization (NMF) is a valuable deconvolution technique for gene expression data, characterized by its non-negativity constraint and ability to deconvolve without prior component knowledge.

Purpose of the Study:

• To introduce FaStaNMF, a novel method designed to enhance the stability, accuracy, and speed of Non-negative matrix factorization (NMF) for gene expression deconvolution.
• To address the challenge of Non-negative matrix factorization (NMF) not guaranteeing globally unique solutions, which impacts reproducibility.
• To provide a computational tool for reproducible and accurate analysis of gene expression data from complex biological samples.

Main Methods:

• Development of FaStaNMF, a Non-negative matrix factorization (NMF) algorithm prioritizing global stability, accuracy, and computational speed.
• Application of FaStaNMF to four distinct datasets with known ground truth, utilizing both publicly available data and the RNAGinesis simulation infrastructure.
• Evaluation of FaStaNMF against standard Non-negative matrix factorization (NMF) approaches for reproducibility.

Main Results:

• FaStaNMF demonstrated favorable comparisons in speed, accuracy, and stability against standard Non-negative matrix factorization (NMF) methods.
• The method achieved enhanced global stability, crucial for inter-experiment and inter-laboratory reproducibility.
• Performance was validated across diverse datasets, confirming its robustness.

Conclusions:

• FaStaNMF offers a significant advancement in computational deconvolution for gene expression analysis.
• The method provides a stable, accurate, and fast approach to extracting cell type-specific expression profiles.
• FaStaNMF is anticipated to be broadly applicable to various biological contexts, including tumor/immune microenvironments and other disease-related samples.