Overview
Microarray data analysis transforms raw fluorescence intensities from hybridized microarrays into meaningful gene expression measurements. Although RNA-seq has largely supplanted microarrays for discovery-based studies, microarrays remain common in clinical diagnostics, plant breeding, and large population studies due to their low cost, standardized protocols, and well-established analysis pipelines. A single microarray can measure expression of tens of thousands of transcripts simultaneously by exploiting complementary base pairing between sample cDNA and immobilized probes. The analysis workflow addresses the unique technical characteristics of microarray data, including background correction, normalization, and probe-level summarization.
Methods
Microarray analysis begins with image processing to extract probe-level intensities. Background correction removes nonspecific signal using methods such as robust multi-array average (RMA) or GC-content adjustment. Normalization makes arrays comparable: quantile normalization is the most common approach for one-color arrays, while loess normalization is applied to two-color designs. Probe summarization (for Affymetrix arrays, using RMA or PLIER) combines multiple probes per gene into a single expression value. Quality assessment uses pseudo-images, NUSE plots, and RLE plots to identify problematic arrays. Differential expression is tested with limma, which uses empirical Bayes moderation to stabilize variance estimates across genes. Batch effects are detected with principal component analysis and corrected using ComBat or limma’s removeBatchEffect.
Applications
Despite the rise of sequencing, microarrays continue to deliver value. The FDA-approved MammaPrint and Oncotype DX tests use microarrays for breast cancer prognosis. Clinical DNA microarrays and gene expression panels guide treatment decisions in oncology and rare disease diagnosis. Microarray data analysis also supports validation of qPCR results through qPCR correlation studies and complements DNA sequencing by profiling expression at scale. In agricultural genomics, microarrays enable cost-effective trait mapping and marker-assisted selection across large breeding populations.
Practical Protocol
For Affymetrix microarray analysis in R/Bioconductor, start by reading CEL files with the oligo or affy package. Create an ExpressionFeatureSet: data <- read.celfiles(list.celfiles()). Perform RMA normalization in one step with eset <- rma(data), which applies background correction (convolution of signal and noise distributions), quantile normalization across arrays, and probe-level summarization via median polish. The resulting expression matrix contains log2-transformed values ready for analysis. Assess array quality with NUSE (Normalized Unscaled Standard Error) and RLE (Relative Log Expression) plots; arrays with median NUSE above 1.05 or RLE interquartile ranges deviating from the norm should be flagged. Differential expression testing uses limma: first fit a linear model with fit <- lmFit(eset, design) where design is a model matrix created with model.matrix(~ condition). Then apply empirical Bayes moderation with fit2 <- eBayes(fit) and extract results with topTable(fit2, coef = 2, number = Inf, adjust = "BH"). The output includes log fold changes, average expression, moderated t-statistics, p-values, and Benjamini-Hochberg adjusted p-values. For two-color arrays, use limma with normalizeWithinArrays (loess normalization) and normalizeBetweenArrays (scale normalization). Batch effects identified in PCA plots can be corrected with ComBat(dat, batch) from the sva package. Visualize results with volcano plots, heatmaps of the top differentially expressed probes, and MA plots to inspect intensity-dependent bias. Annotate probes to genes using the platform-specific annotation package (e.g., hgu133plus2.db for Human Genome U133 Plus 2.0 arrays).
