After count quality assessment and filtering, the RNA-Seq workflow transitions from data preparation to data understanding.
At this stage, expression measurements must be normalized to account for technical differences between samples. Once normalized, exploratory analyses help reveal patterns, relationships, potential outliers, and sources of variation that may influence downstream statistical modeling.
The goal is not yet to identify differentially expressed genes. The goal is to understand the structure of the dataset before formal inference begins.
flowchart TD
A[Filtered Count Matrix]
subgraph EA["Expression Analysis"]
B[Normalization & Exploratory Analysis]
C[Differential Expression Analysis]
end
A --> B --> Cflowchart TD
A[Filtered Count Matrix]
subgraph EA["Expression Analysis"]
B[Normalization & Exploratory Analysis]
C[Differential Expression Analysis]
end
A --> B --> C
This chapter transforms filtered counts into comparable expression measurements and explores the overall structure of the dataset.
RNA-Seq samples often differ in sequencing depth and library composition.
Consider two samples:
| Sample | Total Reads |
|---|---|
| Sample1 | 20,000,000 |
| Sample2 | 40,000,000 |
Even if the underlying biology is identical, Sample2 may contain roughly twice as many observed counts simply because more reads were sequenced.
Without normalization, direct comparisons may be misleading.
Normalization aims to:
Normalization does not remove all variation. It provides a better foundation for downstream analyses.
RNA-Seq workflows commonly use:
Different methods serve different analytical purposes.
DESeq2 estimates sample-specific size factors that account for differences in sequencing depth and library composition.
Conceptually:
Filtered Count Matrix
↓
Estimate Size Factors
↓
Normalized CountsThis approach is widely used in differential expression workflows.
dds <- DESeq2::estimateSizeFactors(dds)
normalized_counts <- DESeq2::counts(
dds,
normalized = TRUE
)The resulting normalized counts can be used for visualization and exploratory analyses.
Exploratory analysis helps answer questions such as:
Exploratory analyses provide context before hypothesis testing begins.
PCA is one of the most widely used exploratory tools in RNA-Seq analysis.
PCA reduces thousands of expression measurements into a small number of components that capture major sources of variation.
Researchers commonly use PCA to:
When reviewing PCA plots, consider:
PCA is an exploratory tool and should not be interpreted as a formal statistical test.
Hierarchical clustering provides another perspective on sample relationships.
Clustering can help identify:
Clustering complements PCA by emphasizing sample similarity.
Heatmaps are frequently used to visualize:
Heatmaps provide a useful visual summary of data structure and sample relationships.
Batch effects are a common source of unwanted variation.
Potential indicators include:
Observed patterns should always be interpreted alongside metadata.
For visualization, transformed expression values are often useful.
Examples include:
vsd <- DESeq2::vst(dds)or
rld <- DESeq2::rlog(dds)These transformations often improve PCA and clustering analyses.
Before proceeding to differential expression analysis, confirm that:
Common mistakes include:
Exploratory analysis should guide understanding and improve subsequent modeling decisions.
This chapter transforms the dataset from filtered counts into an analysis-ready representation.
Filtered Count Matrix
↓
Normalization
↓
Exploratory Analysis
↓
Normalized Expression Data
↓
Differential Expression AnalysisThe output of this stage becomes the primary input for formal statistical modeling.
Normalization improves comparability across samples, while exploratory analysis helps reveal the structure of the dataset.
Together, these steps create a strong foundation for differential expression analysis and help ensure that downstream biological conclusions are based on well-understood data.
The next chapter focuses on differential expression analysis, where normalized expression data are formally modeled to identify genes associated with the biological question.