Skip to content

usage-guide.md

Latest commit

 

History

History
84 lines (62 loc) · 2.65 KB

usage-guide.md

File metadata and controls

84 lines (62 loc) · 2.65 KB
Raw

Statistical Analysis - Usage Guide

Overview

Statistical analysis identifies metabolites associated with biological conditions. Methods range from simple univariate tests to multivariate models for biomarker discovery.

Prerequisites

# R packages
install.packages(c("mixOmics", "ropls", "pROC"))

# Python
pip install scikit-learn scipy statsmodels

Quick Start

Tell your AI agent what you want to do:

  • "Find differentially abundant metabolites between treatment and control"
  • "Run PLS-DA and identify important features by VIP score"

Example Prompts

Univariate Analysis

"Run t-tests comparing treatment vs control with FDR correction" "Perform ANOVA across my three treatment groups and identify significant metabolites" "Calculate fold changes and create a volcano plot"

Multivariate Analysis

"Run PCA for exploratory analysis and check sample grouping" "Build a PLS-DA model with 10-fold cross-validation and calculate VIP scores" "Use OPLS-DA for biomarker discovery between disease and healthy groups"

Biomarker Selection

"Identify metabolites with VIP > 1, FDR < 0.05, and |log2FC| > 1" "Calculate ROC curves and AUC for top candidate biomarkers" "Build a Random Forest classifier and rank feature importance"

Model Validation

"Validate my PLS-DA model with 100 permutation tests" "Report Q2 and R2 from cross-validation"

What the Agent Will Do

  1. Check data distribution and apply appropriate tests
  2. Run univariate tests with multiple testing correction
  3. Build and validate multivariate models
  4. Calculate VIP scores and feature importance
  5. Generate summary tables and visualizations
  6. Export significant metabolites

Tips

  • Always correct for multiple testing (FDR/BH method is standard)
  • Validate PLS-DA with permutation testing (Q2 should exceed permuted values)
  • VIP > 1 is common threshold, but combine with FDR for confidence
  • Use 5-10 fold CV with 50+ repeats for stable model assessment
  • Report both univariate (FDR) and multivariate (VIP) evidence

Method Selection

Method Samples Use Case
t-test 2 groups Simple comparison
ANOVA 3+ groups Multiple conditions
PCA Any Exploratory, QC
PLS-DA 2+ groups Classification, VIP
OPLS-DA 2 groups Biomarker discovery
Random Forest 2+ groups Non-linear importance

Biomarker Criteria

Metric Threshold
FDR < 0.05 (discovery: < 0.1)
log2FC > 1 (2-fold change)
VIP > 1
AUC > 0.7 (moderate), > 0.8 (good)

References

  • mixOmics: doi:10.1371/journal.pcbi.1005752
  • ropls (OPLS-DA): doi:10.1021/acs.jproteome.5b00354