Performance of tumour microenvironment deconvolution methods in breast cancer using single-cell simulated bulk...

Area of Science:

• Computational biology
• Cancer research
• Genomics

Background:

• Tumor microenvironment (TME) cells influence cancer progression and treatment outcomes.
• Bulk RNA-sequencing (RNA-seq) is widely used, but deconvoluting complex TME composition is difficult.
• Single-cell RNA-sequencing (scRNA-seq) offers higher resolution for TME analysis.

Purpose of the Study:

• To evaluate and compare the performance of nine computational methods for deconvolving TME cell composition from bulk RNA-seq data.
• To assess how varying tumor purity and molecular subtypes affect deconvolution accuracy.
• To identify the most robust methods for TME deconvolution, particularly for immune cell populations.

Main Methods:

• Utilized scRNA-seq data from breast tumors to simulate thousands of bulk RNA-seq mixtures.
• Compared the performance of nine deconvolution algorithms: BayesPrism, Scaden, CIBERSORTx, MuSiC, DWLS, hspe, CPM, Bisque, and EPIC.
• Validated findings on two independent breast cancer datasets.

Main Results:

• Method performance varied significantly, with some showing robustness at high tumor purity levels.
• A common misclassification of normal epithelial cells as cancer cells occurred with increasing tumor purity.
• BayesPrism and DWLS demonstrated the lowest false positive and false negative rates, excelling in deconvolving granular immune lineages.
• Breast cancer molecular subtype impacted deconvolution accuracy.

Conclusions:

• Tumor purity is a critical factor influencing TME deconvolution accuracy.
• BayesPrism and DWLS are recommended for TME deconvolution, especially for detailed immune cell profiling.
• Enhanced single-cell characterization of rare cell types is needed.
• Consideration of tumor cell composition is essential for accurate TME deconvolution.