Heterogeneous pseudobulk simulation enables realistic benchmarking of cell-type deconvolution methods.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Computational cell type deconvolution estimates cell type abundance in bulk tissues, crucial for understanding tumor microenvironments.
• Existing benchmarking studies often use simulated pseudobulk datasets generated from randomly selected single cells.
• This standard approach may lack biological variance, impacting the reliability of deconvolution method evaluations.

Purpose of the Study:

• To address the limitations of current bulk simulation pipelines in benchmarking deconvolution methods.
• To propose and validate a heterogeneous simulation strategy for more realistic synthetic bulk datasets.
• To provide a robust framework for evaluating the performance and robustness of various deconvolution algorithms.

Main Methods:

• Developed a heterogeneous simulation strategy to generate synthetic bulk expression data that better reflects biological variance.
• Evaluated the robustness of different classes of deconvolution methods, including reference-free and regression-based approaches.
• Conducted an extensive benchmark of deconvolution methods across eight diverse datasets.

Main Results:

• Standard simulation using random cells generates synthetic data lacking appropriate biological variance.
• Heterogeneously simulated bulk samples exhibit variance matching real bulk datasets, enhancing benchmarking.
• Reference-free deconvolution methods showed poor performance in heterogeneous settings; BayesPrism and a hybrid MuSiC/CIBERSORTx approach were top performers.

Conclusions:

• The proposed heterogeneous simulation method provides a more realistic basis for benchmarking computational cell type deconvolution tools.
• This framework allows for a more nuanced understanding of method performance, particularly concerning biological heterogeneity.
• An open-source package for the simulation method and benchmarking framework is available to facilitate further research and development.