Skip to content

What is DESeq2-inspired Model?

Multi-Sample Session Specific

This feature is only available in multi-sample sessions.

DESeq2-inspired Model: an in-browser, client-side statistical engine in CartoScope designed for real-time differential expression analysis across multi-sample biological cohorts without requiring external R/Bioconductor server backends.

How does the DESeq2-inspired model work?

  • Median-of-Ratios Normalization: Computes geometric means across sample expression profiles and median sample ratios to normalize for library size and sequencing depth differences across slides.
  • Negative Binomial Model: Models sample-level count variances using a Negative Binomial distribution rather than a simple Poisson model.
  • Closed-Form Parametric Dispersion: Uses analytical parametric dispersion estimation to model sample-to-sample variance in JavaScript/Web Worker threads.
  • Wald Z-Score Testing: Calculates Wald test statistics (W = log2 FC / SE) and stable two-tailed p-values.

How is this different from actual R/Bioconductor DESeq2?

CartoScope's deseq2 option is not running the actual R DESeq2 package.

The key differences are:

Feature DESeq2 CartoScope DESeq2-inspired Model
Requirements R runtime environment / server backend. Perform in CartoScope with simple click. No other dependencies.
Model Fitting Fits Generalized Linear Models (GLMs) via iterative maximum likelihood. Uses closed-form analytical equations for real-time calculations.
Dispersion Estimation Fits a smooth trend curve across all genes and performs Empirical Bayes shrinkage. Uses parametric dispersion estimation combined with sample-level variance.
Primary Purpose Final publication-grade statistical modeling in R. Instant interactive web screening, volcano plot exploration, and filtering.

Why use the DESeq2-inspired model over the Chi-squared test?

While Chi-squared (chisq) is fast for screening pooled counts, the DESeq2-inspired model provides robust biological cohort comparisons:

  • Accounts for Biological Variance: Operates at the sample level (pseudo-bulk) to model sample-to-sample variability, preventing false positives from single-slide outliers or batch effects.
  • Robust Normalization: Uses Median-of-Ratios size factor normalization to prevent sequencing depth differences or dominant genes from distorting fold-change (log2 FC) values.
  • Stable Fold Changes: Incorporates pseudocount scaling to avoid artificially inflated fold changes for low-count genes.