ARTdeConv: adaptive regularized tri-factor non-negative matrix factorization for cell type deconvolution

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Accurate cell type deconvolution from bulk gene expression is vital for disease research.
• Existing methods struggle with incomplete signatures, partial information, and varying mRNA amounts, biasing results.
• Limited use of external reference data (e.g., population cell proportions) hinders accuracy.

Purpose of the Study:

• To develop an advanced deconvolution method addressing limitations of current approaches.
• To introduce ARTdeConv (adaptive regularized tri-factor non-negative matrix factorization) for robust cell type deconvolution.
• To validate ARTdeConv's performance against state-of-the-art methods and in real-world applications.

Main Methods:

• Developed an adaptive regularized tri-factor non-negative matrix factorization algorithm (ARTdeConv).
• Established rigorous numerical convergence for the ARTdeConv algorithm.
• Validated performance through benchmark simulations and real-world datasets (influenza vaccine, COVID-19).

Main Results:

• ARTdeConv demonstrated superior performance over existing semi-reference-based and reference-free deconvolution methods.
• The method showed robustness even when its core assumptions were challenged.
• ARTdeConv estimates strongly correlated with flow cytometry measurements in a vaccine study.
• Analysis of COVID-19 patient data revealed immunologically relevant patterns.

Conclusions:

• ARTdeConv offers a significant advancement in cell type deconvolution from gene expression data.
• The R package implementation facilitates its adoption by researchers and practitioners.
• Accurate deconvolution enhances understanding of cellular dynamics in health and disease.