After normalization and exploratory analysis, the next stage of the RNA-Seq workflow is formal statistical inference.
The objective of differential expression analysis is to determine whether observed differences in expression are consistent with the biological question being investigated.
Rather than simply comparing counts, differential expression analysis uses statistical models to evaluate evidence while accounting for biological variability and experimental design.
flowchart TD
A[Normalized Expression Data]
subgraph EA["Expression Analysis"]
B[Differential Expression Analysis]
C[Results QC & Visualization]
end
A --> B --> Cflowchart TD
A[Normalized Expression Data]
subgraph EA["Expression Analysis"]
B[Differential Expression Analysis]
C[Results QC & Visualization]
end
A --> B --> C
This chapter represents the formal statistical modeling stage of the RNA-Seq system.
Differential expression analysis is often used to address questions such as:
The goal is to identify expression differences supported by statistical evidence.
Differential expression analysis evaluates whether observed expression differences are larger than expected by chance.
This requires:
Reliable inference depends on all four components.
RNA-Seq counts are typically modeled using count-based statistical methods.
Popular approaches include:
These methods estimate variation across biological replicates and evaluate evidence for differential expression.
Differential expression models depend on the study design.
A simple design formula might be:
~ conditionA model accounting for batch effects might be:
~ batch + conditionThe design formula links the statistical model to the experimental design and metadata.
dds <- DESeq2::DESeq(dds)Differential expression results can then be extracted.
results_tbl <- DESeq2::results(dds)The resulting table contains statistics used for interpretation.
A differential expression table often contains:
| Gene | log2FC | pvalue | padj |
|---|---|---|---|
| GeneA | 2.1 | 0.0001 | 0.001 |
| GeneB | -1.8 | 0.0008 | 0.004 |
| GeneC | 0.3 | 0.7200 | 0.880 |
Each column provides different information about the observed expression differences.
The log2 fold change describes the magnitude and direction of expression differences.
Examples:
| log2FC | Interpretation |
|---|---|
| +1 | Expression doubled |
| +2 | Expression increased four-fold |
| -1 | Expression reduced by half |
| 0 | No change |
Fold changes describe effect size rather than statistical significance.
P-values evaluate evidence against the null hypothesis.
A small p-value suggests that the observed difference is unlikely to be explained solely by random variation.
However, RNA-Seq studies test thousands of genes simultaneously, which creates a multiple-testing problem.
RNA-Seq experiments often evaluate thousands of genes.
For example:
20,000 genes testedEven if no true differences exist, some genes may appear significant by chance alone.
Multiple-testing correction helps reduce false discoveries.
Most RNA-Seq workflows use adjusted p-values.
Common methods include:
Adjusted p-values help control the expected proportion of false discoveries.
A statistically significant result is not automatically biologically important.
Researchers should consider:
Interpretation requires both statistics and biological reasoning.
Common thresholds include:
Adjusted p-value < 0.05and
|log2 Fold Change| > 1Thresholds should be justified and reported transparently.
Many studies involve multiple possible comparisons.
Examples include:
Treated vs ControlDisease vs HealthyKnockout vs Wild TypeThe chosen contrast should align directly with the biological question.
Large fold changes from low-count genes can be unstable.
Methods such as:
apeglmcan shrink fold-change estimates toward more stable values.
This often improves interpretation and visualization.
Before interpreting results, confirm that:
Common differential expression mistakes include:
These mistakes can weaken downstream biological interpretation.
Differential expression analysis produces statistical results that require quality assessment and visualization.
Normalized Expression Data
↓
Differential Expression Analysis
↓
Statistical Results
↓
Results QC & VisualizationThe next stage focuses on evaluating and communicating these results.
Differential expression analysis transforms normalized expression measurements into statistical evidence.
By combining study design, biological replication, normalization, and statistical modeling, researchers can identify genes associated with biological conditions while accounting for variability and uncertainty.
The next chapter focuses on results quality assessment and visualization, where statistical findings are evaluated, summarized, and prepared for biological interpretation.