Benchmarking transcriptome deconvolution methods for estimating tissue- and cell-type-specific extracellular vesicle...

Area of Science:

• Biochemistry
• Molecular Biology
• Bioinformatics

Background:

• Extracellular vesicles (EVs) carry crucial biological molecules, but quantifying their tissue- and cell-type-specific origins in biofluids is difficult.
• Deconvolution methods, which analyze transcriptomes, are promising for estimating EV abundance but require performance evaluation on EV data.

Purpose of the Study:

• To benchmark and compare the performance of 11 deconvolution methods for estimating tissue- and cell-type-specific EV abundances in human biofluids.
• To identify factors influencing deconvolution accuracy and assess method concordance across different sample types and isolation procedures.

Main Methods:

• Evaluated 11 deconvolution algorithms using cell line EVs, in silico mixtures, human plasma EVs (n=118), and urine EVs (n=88).
• Assessed method accuracy in estimating cell-type-specific abundances from pure and mixed EV samples.
• Compared results from four deconvolution methods across two urine EV cohorts and evaluated plasma EV estimates.

Main Results:

• Identified specific deconvolution methods with high accuracy for pure and in silico mixed EV samples.
• Demonstrated high concordance among four deconvolution methods for urine EVs, irrespective of isolation techniques.
• Showed strong agreement among three methods for plasma EV abundance estimates and identified key factors for accuracy.

Conclusions:

• The deconvolution algorithms DWLS and CIBERSORTx provide accurate and concordant estimates of tissue- and cell-type-specific EV abundances in biological fluids.
• Incorporating biological knowledge into signature creation is vital for improving deconvolution accuracy.
• This work validates deconvolution approaches for characterizing EV populations in biofluids.