SCENT: Single Cell ENTropy for Differentiation Potency Estimation

Overview

SCENT is an R package for estimating differentiation potency of single cells from scRNA-Seq data using signaling entropy on protein-protein interaction (PPI) networks. The package implements the computational framework described in Teschendorff & Enver (2017), with performance optimizations through C++ (Rcpp/RcppArmadillo).

📖 Documentation: https://zaoqu-liu.github.io/SCENT/

Theoretical Background

Signaling Entropy Rate (SR)

The differentiation potency of a cell can be quantified by the signaling entropy rate, which measures the "randomness" of information flow through the protein interaction network. The mathematical formulation is:

$$SR = \frac{1}{SR_{max}} \sum_{j} \pi_j S_j$$

where:

$\pi_j$: Stationary distribution at gene $j$, computed as $\pi_j = \frac{x_j \cdot (\mathbf{A}\mathbf{x})_j}{Z}$
$S_j$: Local entropy at gene $j$, defined as $S_j = -\sum_k P_{jk} \log P_{jk}$
$P_{jk}$: Transition probability from gene $j$ to $k$, given by $P_{jk} = \frac{A_{jk} \cdot x_k}{\sum_l A_{jl} \cdot x_l}$
$SR_{max}$: Maximum entropy rate, equal to $\log(\lambda_1)$ where $\lambda_1$ is the largest eigenvalue of $\mathbf{A}$

Biological Interpretation

Cell State	SR Value	Network Signaling
Pluripotent (stem-like)	High	Uniform flow through multiple pathways
Differentiated	Low	Concentrated flow through specific pathways

Installation

From R-Universe (Recommended)

install.packages("SCENT", repos = c(
  "https://zaoqu-liu.r-universe.dev",
  "https://cloud.r-project.org"
))

From GitHub

# Install dependencies
install.packages(c("Rcpp", "RcppArmadillo", "Matrix", "igraph"))

# Install SCENT
devtools::install_github("Zaoqu-Liu/SCENT")

System Requirements

Platform	Requirement
macOS	`xcode-select --install`
Windows	Rtools
Linux	Standard build tools (gcc, g++)

Quick Start

library(SCENT)

# Load built-in PPI network (8,434 genes, 303,600 interactions)
data(net13Jun12.m)

# Load your expression matrix (genes × cells)
# For Seurat objects: exp <- GetAssayData(seurat_obj, layer = "data")

# Method 1: CCAT (fast approximation)
ccat_scores <- CompCCAT(exp, net13Jun12.m)

# Method 2: SR (full computation)
integ <- DoIntegPPI(exp, net13Jun12.m)
sr_result <- CompSRana(integ, local = TRUE)

# Access results
sr_scores <- sr_result$SR           # Global entropy rate per cell
local_entropy <- sr_result$locS     # Gene-level local entropy
stationary <- sr_result$inv         # Stationary distribution

Methods Comparison

Method	Algorithm	Speed	Accuracy	Use Case
SR	Full entropy computation	~5s / 100 cells	Gold standard	Publication, validation
CCAT	Correlation with degree	~0.1s / 100 cells	Approximation (r ≈ 0.78)	Screening, large datasets

Tutorials

Tutorial	Description
Quick Start	Basic usage and workflow
Algorithm	Mathematical background and derivation
Visualization	Publication-quality figures
Benchmark	Performance analysis

Citation

If you use SCENT in your research, please cite:

Teschendorff AE, Enver T. Single-cell entropy for accurate estimation of differentiation potency from a cell's transcriptome. Nature Communications. 2017;8:15599. doi:10.1038/ncomms15599

@article{teschendorff2017single,
  title={Single-cell entropy for accurate estimation of differentiation 
         potency from a cell's transcriptome},
  author={Teschendorff, Andrew E and Enver, Tariq},
  journal={Nature Communications},
  volume={8},
  pages={15599},
  year={2017},
  doi={10.1038/ncomms15599}
}

Authors

Zaoqu Liu — Maintainer, C++ optimization (GitHub · ORCID)
Andrew E Teschendorff — Original algorithm developer