Estimating the effect of tissue- and blood-derived cell reference matrices on deconvolving bulk transcriptomic...

Area of Science:

• Computational Biology
• Transcriptomics
• Immunology

Background:

• Cell deconvolution analyzes mixed cell populations in bulk transcriptomic data.
• The source of cell reference matrices (CRMs) for deconvolution is critical but understudied.
• Existing CRMs are often derived from blood, potentially limiting tissue-specific analyses.

Purpose of the Study:

• To systematically evaluate the impact of tissue- vs. blood-derived CRMs on cell deconvolution accuracy.
• To develop and benchmark novel tissue- and blood-derived CRMs using inflammatory bowel disease (IBD) scRNA-seq data.
• To assess CRM performance in both IBD and lung adenocarcinoma (LUAD) datasets.

Main Methods:

• Developed custom tissue- and blood-derived CRMs from IBD scRNA-seq data.
• Benchmarked custom CRMs against public CRMs (IRIS, LM22, ImmunoStates).
• Evaluated deconvolution performance using public bulk transcriptomic datasets, simulated samples, and LUAD data via goodness-of-fit and cell fraction correlation.

Main Results:

• Tissue-derived CRMs significantly outperformed blood-derived CRMs for bulk tissue transcriptome deconvolution.
• Tissue-derived CRMs showed higher goodness-of-fit and more accurate cell proportion estimates, especially for immune and stromal cells.
• Performance differences diminished when deconvolving bulk blood transcriptomics; similar trends observed in LUAD datasets.

Conclusions:

• Selecting appropriate CRMs is vital for accurate cell deconvolution in bulk tissue samples.
• Tissue-derived CRMs are superior for analyzing tissue-specific cellular compositions in fields like immunology and oncology.
• The R package DeconvRef facilitates the creation of user-defined CRMs for broader research applications.