RNA-Seq Count Model Explorer — Poisson vs Negative Binomial (NB2) + Streamlit Live Demo

An interactive bioinformatics/statistical modeling project that explores RNA-seq count data and compares Poisson vs Negative Binomial (NB2) assumptions. The Streamlit app visualizes overdispersion (mean–variance behavior), fits per-gene GLMs, and exports model comparison results.

Live Demo: LIVE_DEMO_URL_HERE
GitHub Repo: REPO_URL_HERE

---

Why this matters

RNA-seq data are counts, and Poisson models assume:

Var(Y) = E(Y)

In practice, RNA-seq often shows overdispersion (variance > mean), which can make Poisson a poor fit. Negative Binomial (NB2) models handle this by introducing a dispersion term:

Var(Y) = μ + α μ²

This project makes that distinction easy to see and test interactively.

---

Features

Streamlit App

• Three data modes

  • Sample dataset (runs instantly)
  • Simulated RNA-seq dataset (interactive controls)
  • Upload your own CSV in tidy format

• Simulation controls

  • Number of genes / samples
  • Dispersion α (controls overdispersion)
  • Condition effect strength
  • Random seed

• Model comparison

  • Gene-wise Mean vs Variance visualization
  • Per-gene intercept-only GLMs:
    • Poisson GLM (baseline)
    • NB2 GLM (gene-wise α estimated via method-of-moments)
  • AIC comparison plot and per-gene results table

• Export

  • Download per-gene model comparison results as CSV

---

Expected CSV format (Upload mode)

Your uploaded CSV should be tidy with:

Required columns:

• gene
• sample
• count

Optional column:

• condition

Example:

gene,sample,condition,count
GENE1,S1,Control,12
GENE1,S2,Control,15
GENE1,S3,Treated,40