For two decades, microarrays were the standard for measuring gene expression. Then RNA-seq arrived, and within a few years, the literature flipped. RNA-seq is now the default — but microarrays still earn their keep in specific contexts.
How they work, briefly
Microarrays: A glass slide carries thousands to millions of probe sequences (oligonucleotides) at known positions. Labeled cDNA or cRNA is hybridized to the array, washed, and scanned. Fluorescent signal at each probe quantifies transcript abundance.
RNA-seq: RNA is converted to cDNA, fragmented, ligated to adapters, sequenced on a next-gen platform, and aligned to a reference transcriptome. Counts of reads mapped to each transcript estimate expression.
Why RNA-seq won
- No reference required. Microarrays only detect transcripts whose sequence is on the array. RNA-seq discovers novel transcripts, splice junctions, and even unknown organisms.
- Higher dynamic range. Microarray signals saturate at high abundance and disappear into noise at low abundance. RNA-seq spans 5–6 orders of magnitude.
- Single-base resolution. RNA-seq resolves splicing, allele-specific expression, and SNPs.
- Cross-species applicability. Same workflow for any organism with a reference (or de novo).
Side-by-side comparison
| Property | Microarray | RNA-seq |
|---|---|---|
| Reference required | Yes (probe design) | Useful but not strict |
| Dynamic range | ~3 orders of magnitude | ~5–6 orders of magnitude |
| Detects novel transcripts | No | Yes |
| Resolves splicing | Limited | Yes |
| Cost per sample | Low–Moderate | Higher |
| Bioinformatics burden | Lower | Higher |
| RNA input | ~100 ng+ | Down to single cell |
When microarrays still make sense
- Clinical assays where regulatory approval already exists for specific arrays
- Genotyping arrays for GWAS — millions of SNPs measured cheaply
- Established legacy workflows with extensive historical comparison data
- Resource-limited settings where bioinformatics infrastructure isn’t available
Common pitfalls when transitioning
- Don’t directly compare microarray fold-changes to RNA-seq fold-changes — the metrics aren’t equivalent
- Batch effects matter even more in RNA-seq because library prep introduces additional variance
- RNA-seq needs more samples than you might think. Three replicates per condition is a minimum
What about NanoString and nCounter?
For targeted gene panels (50–800 genes), NanoString sits between microarrays and RNA-seq: hybridization-based, no amplification, very reproducible. Useful for clinical or focused research applications where you know exactly which transcripts you care about.
RNA-seq is the default in modern research. Microarrays still survive in clinical genotyping, validated diagnostic panels, and cost-constrained discovery work.
