MaGIC Differential Expression Tool

Welcome to the Differential Expression Tool by the Molecular and Genomics Informatics Core (MaGIC).

What This Tool Does

This is the DE engine of the MaGIC bulk-expression pipeline. Upload a raw count matrix and sample metadata, build a design, and run DESeq2, edgeR-GLM, or limma-voom. It produces a results table per contrast — gene, baseMean, log2FoldChange, lfcSE, stat, pvalue, padj — that drops straight into magic-volcano, magic-heatmap, and magic-setcomparison.

Which DE method should I use?

DESeq2

Small-to-moderate sample sizes (~3–10 per group). Shrunken fold-change estimates and the most permissive about low counts. The default choice for most experiments.

edgeR-GLM

Flexible designs, robust to complex multi-factor models, with power comparable to DESeq2 in most settings.

limma-voom

Larger sample sizes (>10 per group). Fast and well-calibrated when you trust the mean-variance assumption.

All three normalize raw counts internally — DESeq2 uses median-of-ratios; edgeR and limma-voom use TMM. Feed this tool raw integer counts, not the normalized output of magic-qc.

When should I correct for batch?

First-line: add batch as a covariate in your design

In Setup & Run → Design, add your batch variable as a covariate. The model handles batch within the statistical framework — no counts modified, fewer assumptions, no inflated Type I error. Try this first for any DE analysis.

Alternative: pre-correct counts with ComBat-seq

Pre-correcting the counts (Setup & Run → Batch Correction) is appropriate only when batch is severely confounded with biology, when the covariate approach produces unstable fits, or when you need a batch-corrected count matrix as an output artifact for visualization / clustering tools that cannot model batch themselves. ComBat-seq discards information and can inflate false positives if used unnecessarily — don't reach for it by default.

Required Input Data Formats

Raw Count Matrix
  • File format: CSV or TSV
  • Rows: genes (one gene per row)
  • Columns: samples (one sample per column)
  • First column: gene identifiers
  • Values: raw (un-normalized) integer counts
GeneID, Control1, Control2, Treat1
G0001,  149,      122,      218
G0002,  409,      151,      46
Sample Metadata
  • File format: CSV or TSV
  • Rows: samples (one sample per row)
  • First column: sample names — must match matrix column names
  • Additional columns: condition, batch, sex, time, ...
Sample,    Condition, Batch
Control1,  Control,   B1
Treat1,    TreatA,    B2
Gene Info Table (optional)

Maps gene IDs to symbols so the results tables carry a Symbol column for downstream annotation.

GeneID,  Symbol
G0001,   TP53
G0002,   EGFR
1 · Data

Demo data is loaded by default: 600 genes × 12 samples in 3 conditions (Control / TreatA / TreatB), with a partially confounded batch and ~100 truly DE genes per treatment. Configure the design below and click Run.

Flip the switch above to upload your own counts + metadata.
2 · Design
Fold-changes are reported relative to the reference level.
Added to the model additively (no interactions in v1). Add batch here as the first-line way to correct for it.
Batch Correction (ComBat-seq)
Prefer adding batch as a covariate above. Use ComBat-seq only when batch is severely confounded with biology, when covariate fits are unstable, or when you need a corrected count matrix as an output artifact.
Passed to ComBat-seq's group / covar_mod so real biological differences are not removed.
3 · Run
Wald tests each contrast directly. LRT tests the condition term as a whole (reduced model = covariates only).
Quasi-likelihood F-tests (glmQLFit / glmQLFTest) on a filtered count matrix.
voomWithQualityWeights down-weights lower-quality samples. TMM is always applied for limma-voom.

Design formula


                      

                      
Contrasts to test

                      
By default, every non-reference level of the primary condition is compared against the reference. Add custom level-vs-level comparisons below.

                      

                        

                          

                            
                            

                              
                              
                            

                          

                        

                        

                          

                            
                            

                              
                              
                            

                          

                        

                        

                          

                          
                        

                      

                      
                      

                      
                    

                    

                      

                      
Replicates per level of the primary condition — a quick check on the power available for each contrast.

                      
                    

                    

                      

                      
                    

                    

                      

                      
                    

                    

                      

                      
                    

                    

                      

                      
                    

                  

                

              

            

          

          

            

              

                

                  
Results Options

                  

                  
Significance thresholds

                  

                    Each contrast tab has its own padj and |log2FC| sliders driving its live 'n significant' badge and filtered download.
                  

                  

                  
Table columns

                  

                    
                    

                      

                        
                      

                      

                        
                      

                      

                        
                      

                      

                        
                      

                      

                        
                      

                      

                        
                      

                      

                        
                      

                      

                        
                      

                    

                  

                  

                  
Download

                  

                    
                    

                      
                      
                    

                  

                  

                

              

              

                
              

            

          

          

            

              

                

                  
Diagnostics

                  

                  

                    
Contrast

                    

                      
                      

                        
                        
                      

                    

                    

                  

                  

                    
Fonts

                    

                      
                      
                    

                    

                      
                      
                    

                    

                      
                      
                    

                    

                      
                      
                    

                    

                      
                      
                    

                    

                      

                      
Legend

                      

                        
                        

                          
                          
                        

                      

                      

                        
                        
                      

                      

                        
                        
                      

                    

                  

                  

                    

                    
MA highlight

                    

                      
                      
                    

                    

                      
                      
                    

                    

                      
                      
                    

                  

                  

                    

                    
Histogram

                    

                      
                      
                    

                  

                  

                    
Styling

                    

                      
                      
                    

                    This plot is drawn for the differential expression method you ran.
                  

                  

                  
Size

                  

                    

                      
                      
                      
                    

                  

                  

                    

                      
                      
                    

                    

                      
                      
                    

                  

                

              

              

                

                  
                  

                    

                      

                      
                      

                        

                          

                            
                            

                              
                              
                            

                          

                        

                        

                          
                            
                            Download the MA plot
                          
                        

                      

                    

                    

                      

                      
                      

                        

                          

                            
                            

                              
                              
                            

                          

                        

                        

                          
                            
                            Download the p-value histogram
                          
                        

                      

                    

                    

                      

                      

                      
                      

                        

                          

                            
                            

                              
                              
                            

                          

                        

                        

                          
                            
                            Download the plot
                          
                        

                      

                    

                    

                      

                      

                      
                      

                        

                          

                            
                            

                              
                              
                            

                          

                        

                        

                          
                            
                            Download the PCA plot