tools-only mixOmics Analysis - Usage Guide
Overview
mixOmics provides multivariate methods for multi-omics integration, including both supervised (DIABLO, sPLS-DA) and unsupervised (sPLS, sPCA) approaches with sparse feature selection.
Prerequisites
BiocManager::install('mixOmics')
Quick Start
Tell your AI agent what you want to do:
- "Classify my samples using RNA and protein data with DIABLO"
- "Find correlated features between transcriptomics and metabolomics"
Example Prompts
Classification
"Use DIABLO to classify tumor vs normal samples using my RNA-seq and proteomics data"
"Run sPLS-DA on my expression matrix to identify biomarkers distinguishing treatment groups"
Feature Selection
"Find the most discriminative features across my three omics layers for predicting response"
"Select sparse biomarker panels from each omics view using DIABLO"
Correlation Analysis
"Identify genes and metabolites that co-vary across my samples using sPLS"
"Find correlated protein-metabolite pairs driving the separation between groups"
Multi-Study
"Combine RNA-seq from three studies using MINT to find robust biomarkers"
What the Agent Will Do
- Load and preprocess multi-omics matrices
- Select appropriate method (sPLS, sPLS-DA, DIABLO, MINT)
- Tune keepX parameters using cross-validation
- Build the integration model
- Extract selected features per component
- Generate visualization plots (sample plots, loading plots, circos)
- Assess model performance if supervised
Method Selection Guide
| Method | Supervised | Blocks | Use Case |
|---|---|---|---|
| sPCA | No | 1 | Dimension reduction |
| sPLS | No | 2 | Find correlated features |
| sPLS-DA | Yes | 1 | Classify with feature selection |
| DIABLO | Yes | 2+ | Multi-omics classification |
| MINT | Yes | 1 | Multi-study integration |
Tips
- Use MOFA2 for unsupervised discovery; mixOmics excels at supervised classification
- Tune keepX (features per component) with cross-validation for best performance
- ncomp usually 2-5; check classification performance to choose
- For DIABLO, design matrix controls correlation structure between blocks (0=independent, 1=correlated)
- Pre-filter low-variance features and log-transform count data before analysis
References
- mixOmics: doi:10.1371/journal.pcbi.1005752
- DIABLO: doi:10.1093/bioinformatics/bty1054
- Website: http://mixomics.org/