immunedeconv

immunedeconv provides unified access and evaluation of computational methods to estimate immune cell fractions from bulk RNA-seq data for analysis of tumor microenvironment composition.

Key Features:

• Unified Interface: Integrates multiple computational deconvolution methods, including CIBERSORT, EPIC, quanTIseq, TIMER, xCell, and MCPcounter.
• Systematic Benchmarking: Evaluates accuracy of cell-type quantification across nine immune- and stromal cell types using both simulated and real-world datasets.
• Simulation-based Validation: Uses a single-cell RNA-seq dataset of approximately 11,000 tumor microenvironment cells to simulate bulk samples with known cell-type proportions for validation against gold-standard estimates.
• Extensive Evaluation: Condenses over a hundred thousand predictions to assess performance across seven computational methods, nine cell types, and roughly 1,800 samples from five datasets.
• Guidance for Future Research: Highlights the need to refine cell population definitions and develop reliable or fuzzy cell-type signatures to improve deconvolution accuracy.
• Reproducibility: Provides a Snakemake pipeline for reproducing the benchmark analyses at https://github.com/grst/immune_deconvolution_benchmark.

Scientific Applications:

• Immuno-oncology: Quantifying immune-cell composition in the tumor microenvironment to study tumor progression and immune interactions.
• Prognostic and Predictive Biomarkers: Informing prediction of disease progression and treatment outcomes based on inferred immune-cell fractions.
• Method Selection and Benchmarking: Enabling comparative evaluation and selection of deconvolution methods for transcriptomic studies and biomarker development.

Methodology:

Integrates deconvolution methods (CIBERSORT, EPIC, quanTIseq, TIMER, xCell, MCPcounter), simulates bulk RNA-seq from a ~11,000-cell single-cell RNA-seq TME dataset with known proportions, benchmarks across nine immune- and stromal cell types using simulated and real datasets, condenses >100,000 predictions across seven methods and ~1,800 samples from five datasets, and uses a Snakemake pipeline (https://github.com/grst/immune_deconvolution_benchmark) to reproduce benchmark analyses.