deconvolution estimate immune cell subsets

deconvolution estimate immune cell subsets quantifies immune cell subset proportions from bulk gene expression profiles to enable precise profiling of immune composition in complex tissues such as tumors and peripheral blood mononuclear cells (PBMCs).

Key Features:

• Core algorithm: Employs ε-support vector regression (ε-SVR) incorporating an L1-norm penalty in the loss function.
• Reference signature matrix: Built from differentially expressed genes selected from 148 microarray-based profiles across nine immune cell types.
• Feature selection: Uses ANOVA and minimizes condition number during gene selection to improve matrix robustness.
• Balanced replicates: Utilizes more balanced microarray replicates for key immune cell types to reduce profiling bias.
• Targeted immune subsets: Explicitly profiles T helper cells, regulatory T cells, and macrophage M1/M2 subsets.
• Benchmarking: Compared directly with CIBERSORT using in silico and real-sample benchmarks, showing improved performance in complex mixtures.
• Validation datasets: Includes in silico breast tissue–immune cell mixtures at 30:70 and 50:50 proportions and real human PBMC samples (n=164) for performance assessment.

Scientific Applications:

• Immune composition profiling: Quantification of immune cell subset proportions in tumor tissues and PBMCs.
• Immune-tumor interaction studies: Investigation of pathogenic roles of specific immune cells (e.g., T helper, regulatory T cells, M1/M2 macrophages) within complex tissues.
• Comparative method evaluation: Benchmarking and performance comparison against existing deconvolution approaches such as CIBERSORT.

Methodology:

Constructs a reference signature matrix from DE genes selected via ANOVA and condition-number minimization across 148 microarray profiles of nine immune cell types, then applies ε-SVR with an L1-norm penalty for deconvolution; performance validated against CIBERSORT using in silico breast tissue–immune mixtures (30:70, 50:50) and PBMC samples (n=164).