Skip to content

usage-guide.md

Latest commit

 

History

History
74 lines (53 loc) · 2.51 KB

usage-guide.md

File metadata and controls

74 lines (53 loc) · 2.51 KB

Differential Abundance - Usage Guide

Overview

Identify proteins with significantly different abundance between experimental conditions using statistical testing and multiple testing correction.

Prerequisites

pip install numpy pandas scipy statsmodels
# R packages: BiocManager::install(c("limma", "MSstats", "DEP", "proDA", "QFeatures"))

Quick Start

Tell your AI agent what you want to do:

  • "Find differentially abundant proteins between treatment and control"
  • "Run limma analysis on my normalized protein matrix"
  • "Generate a volcano plot showing significant proteins"

Example Prompts

Statistical Testing

"Run limma differential analysis comparing treatment vs control groups"

"Perform a t-test on each protein with Benjamini-Hochberg correction"

"Use MSstats to test for differential abundance in my label-free data"

Complex Designs

"Set up a limma model with treatment and batch as covariates"

"Test for interaction effects between drug and timepoint"

"Run paired differential analysis for my before/after samples"

Results Filtering

"Filter to proteins with adjusted p-value < 0.05 and |log2FC| > 1"

"Extract the top 50 most significantly changed proteins"

"Identify proteins changing in the same direction across all comparisons"

Visualization

"Create a volcano plot highlighting the significant proteins"

"Make a heatmap of the top differentially abundant proteins"

What the Agent Will Do

  1. Load normalized, imputed protein matrix
  2. Define experimental design and contrasts
  3. Fit statistical model (limma/t-test/MSstats)
  4. Apply multiple testing correction (BH FDR)
  5. Filter by significance thresholds
  6. Generate results table and visualizations

Statistical Methods

Method Description Best For
t-test Simple two-group comparison Large sample sizes
limma Empirical Bayes moderated t-test Small sample sizes
MSstats Mixed-effects models Complex designs
proDA Probabilistic dropout model High missing values

Significance Thresholds

Typical thresholds for proteomics:

  • Adjusted p-value: < 0.05 (or 0.01 for stringent)
  • Log2 fold change: > 1 (2-fold) or > 0.58 (1.5-fold)

Tips

  • Always use adjusted p-values (not raw p-values)
  • Include batch as covariate if samples were processed separately
  • Need n >= 3 replicates per group for reliable statistics
  • Check volcano plot for bias (symmetric up/down regulation expected)
  • Report number tested, thresholds, and number significant