Skip to content

README.md

Latest commit

 

History

History
66 lines (51 loc) · 2.09 KB

README.md

File metadata and controls

66 lines (51 loc) · 2.09 KB
Raw

dCAM

We present reference-free differential Convex Analysis of Mixtures (dCAM), a joint deconvolution method of analyzing grouped bulk data of varying composition that can improve detection of cell type specific differential expressions in many biological contexts. With an L1-norm differential expression penalty-based regularization, dCAM jointly and iteratively estimates both cell type proportions and cell type specific gene expressions in grouped bulk data, and performs cell type-specific differential expression analysis between groups. Through realistic grouped bulk simulations, we demonstrate that dCAM can better detect differentially expressed genes than existing methods. We also report real-data case studies to validate the applicability of dCAM in biomedical research.

Installation

You can install dCAM using devtools:

install.packages("devtools")
pak::pkg_install("cbil-vt/dCAM")

Example

This is a basic example which shows the usage of dCAM. The synthetic data constains 3 cell types, 7799 genes, and 20 sample (10 in group 1, and 10 in group 2). We simulated a small portion of down- or up-regulated genes based on a sorted cell bulk RNA-seq data (GSE73721). This examples should take about 3 minutes to finish.

library(dCAM)

# A synthetic data
data(dataset)

# Run dCAM
fit <- dCAM::run_dcam(
  dataset,
  K = 3,
  cam_params = list(radius.thres = 0.96),
  reweight = "once",
  freeze_A = FALSE,
  max_iter = 5
)

# Performance evalution using pAUC
up <- compute_deg_metrics(fit$delta_S, dataset$truth$is_up,   0.05, "up")
dn <- compute_deg_metrics(fit$delta_S, dataset$truth$is_down, 0.05, "down")
cat(sprintf("  detection (pAUC):          up=%.4f  down=%.4f\n",
            up$pooled_pauc_norm, dn$pooled_pauc_norm))

In this example, for up-regulated genes, we have pAUC of 0.847, while for down-regulated genes, the pAUC is 0.347. This performance difference is expected, as down regulation leads to lower signal to noise ratio and is more challenging to detect.