What Is a Microarray Test and How Does It Work?

A microarray test is a laboratory method that screens thousands to millions of biological targets at once by arranging tiny probes on a solid surface, usually a glass slide or silicon chip. Think of it as a highly organized checklist: each spot on the chip is designed to bind to one specific stretch of DNA, RNA, or protein, and the pattern of binding that lights up tells researchers or clinicians what is present in a sample. Originally developed in the mid-1990s to measure which genes are active in a cell, microarray technology has since expanded into prenatal diagnosis, cancer profiling, infectious disease screening, and beyond.

How a Microarray Actually Works

The core idea behind every microarray is molecular recognition. Short sequences of single-stranded DNA or RNA (called probes) are fixed to precise locations on a chip. When a prepared sample is washed over the surface, any matching molecules in the sample bind to their complementary probes, a process called hybridization. Before the sample is applied, its molecules are tagged with fluorescent dyes. After hybridization, a laser scanner reads the chip and measures how brightly each spot glows. Bright spots mean a lot of the target was present; dim or dark spots mean little or none was there.

The miniaturization itself matters. Research on how spot size affects capture efficiency has shown that shrinking the probe spots leads to more uniform target coverage and faster hybridization, which boosts the signal and speeds up the assay.¹ That is why modern arrays pack hundreds of thousands of probe spots onto a chip the size of a microscope slide. Each spot occupies an area smaller than the tip of a needle.

Chips are manufactured using either contact printing, where tiny pins stamp probe material onto the surface, or non-contact methods like inkjet-style deposition and photochemistry-based techniques. Contact printing is the older approach, valued for reproducible spot formation and straightforward maintenance but limited in throughput. Non-contact methods are newer and more varied, allowing finer features and faster production runs.²

Reading the Results With One-Color and Two-Color Designs

Early gene expression microarrays used a two-color design: one sample was labeled with a green dye and a reference sample with red, then both were applied to the same chip. Spots that lit up green meant that gene was more active in the test sample; red meant more active in the reference. Yellow meant roughly equal. This side-by-side comparison on a single chip was intuitive but required careful balancing of the two dyes.

One-color designs came later. Each sample gets its own chip and one dye, and comparisons happen computationally afterward. A large-scale quality-control project found that the two approaches produce essentially equivalent data quality, with high correlation and strong agreement on which genes are differentially expressed.³ A follow-up study comparing the two designs in clinical outcome prediction for neuroblastoma patients confirmed that neither approach outperformed the other in a meaningful way.⁴ Today, one-color systems dominate commercial platforms because they simplify the workflow and make it easier to compare data across different experiments.

Gene Expression Arrays

The first and still most widely recognized application of microarrays is measuring gene expression, essentially asking which genes are turned on or off in a given tissue. A pioneering 1995 study demonstrated that complementary DNA (cDNA) printed robotically onto glass slides could quantitatively track expression of many genes simultaneously.⁵ Subsequent work assessed how reliably these cDNA platforms captured real changes in messenger RNA levels and found that, when optimized, they could generate accurate and reproducible data suitable for building gene expression databases.⁶

Oligonucleotide arrays, which use shorter synthetic probes instead of full-length cDNA clones, became the commercial standard with platforms from companies like Affymetrix and Illumina. When researchers compared oligonucleotide array measurements to the gold-standard technique of quantitative real-time PCR, the correlation was strong, around 0.89 when targeting the same transcripts.⁷ That level of agreement made gene expression arrays a trusted workhorse in research labs throughout the 2000s and into the 2010s.

SNP Arrays and Copy Number Detection

Not all microarrays measure gene activity. Single nucleotide polymorphism (SNP) arrays are designed to detect specific genetic variants across the genome. Each probe targets a known position where human DNA commonly varies by a single letter. By scanning hundreds of thousands of these positions, a SNP array builds a genome-wide profile of an individual’s genetic makeup.

SNP arrays are particularly useful for finding copy number variations, stretches of DNA that have been duplicated or deleted. These duplications and deletions are widespread in human genomes and can have real functional consequences, including changes in how much protein a gene produces or the creation of abnormal gene fusions.⁸ SNP arrays can also identify a phenomenon called loss of heterozygosity, where both copies of a chromosomal region come from the same parent instead of one from each. This pattern shows up in certain cancers and in some prenatal conditions.⁹ In a large prenatal study using SNP arrays, loss of heterozygosity was detected in about 1% of fetuses, with most cases involving a single chromosome.¹⁰

Prenatal and Developmental Diagnosis

Chromosomal microarray analysis (CMA) has become the go-to genetic test for evaluating children with unexplained developmental delay, intellectual disability, or autism spectrum disorder, and it plays an increasingly important role in prenatal care. A consensus statement from genetics professional organizations declared CMA a first-tier clinical diagnostic test for individuals with developmental disabilities or congenital anomalies, replacing the older technique of karyotyping as the initial screen.¹¹

The diagnostic yield, meaning the percentage of patients who get a clinically meaningful answer, varies by condition. In a study of 410 Han Chinese patients with autism or intellectual disability, CMA identified a pathogenic or likely pathogenic finding in about 21% of cases overall. The yield was substantially higher for intellectual disability and developmental delay (roughly 32%) than for autism alone (about 3%).¹² A cohort of 300 Indian patients found a diagnostic yield of 16% for CMA, which was 9 percentage points higher than traditional karyotyping on the same patients.¹³

In prenatal settings, CMA picks up clinically significant deletions or duplications in about 1% of structurally normal pregnancies and around 6% of pregnancies where an ultrasound has already shown a structural abnormality.¹⁴ A study focused on third-trimester abnormalities found an overall CMA diagnostic yield of about 6%, rising to over 16% when multiple fetal malformations were present. The same study estimated that standard noninvasive prenatal testing (NIPT) could potentially miss up to 70% of the significant findings that CMA caught, since NIPT primarily screens for whole-chromosome conditions like trisomies.¹⁵ Fetuses with heart defects seem to benefit especially from CMA: one single-center study found a 12% diagnostic yield in cardiovascular anomalies, the highest among the organ systems examined.¹⁶

Cancer and Gene Expression Profiling

Microarrays changed how researchers and clinicians think about cancer classification. Instead of grouping tumors solely by where they grow or how they look under a microscope, gene expression profiling can sort them by which genes are active, revealing molecular subtypes that behave very differently even though they appear similar pathologically.

Breast cancer has been the proving ground. Early cDNA microarray studies of nearly 100 patients identified gene signatures that split tumors into distinct subgroups with different outcomes. Genes involved in cell division and chromosomal stability were consistently elevated in the poor-prognosis groups.¹⁷ This line of research led to commercially available tests like Oncotype DX and MammaPrint, which use gene expression data to predict whether a given patient is likely to benefit from chemotherapy or can safely skip it.¹⁸ These are among the best-validated examples of microarray-derived diagnostics guiding real treatment decisions today. The molecular signatures they detect help reveal the biological spectrum of breast cancers and provide both prognostic and predictive information.¹⁹

Screening for Antibiotic Resistance and Infectious Agents

Microarrays have found a niche in microbiology that most people never hear about: screening bacteria for antibiotic resistance genes. One platform was developed to detect up to 90 resistance genes in gram-positive bacteria in a single hybridization step. The disposable chip format offered rapid screening with potential applications in food safety, basic research, and resistance surveillance programs.²⁰ Another custom microarray was designed to target nearly 3,000 antibiotic resistance genes in the human gut microbiome, covering 369 resistance gene types, and used it to show that the diversity of resistance genes changes with age.²¹

The ambition has grown. A more recent assay screens simultaneously for 44 sepsis-relevant pathogen species, 360 virulence factors, and 409 antibiotic resistance genes on a single microarray chip.²² For a patient with a bloodstream infection, that kind of comprehensive snapshot could, in principle, tell clinicians both what organism they are dealing with and which drugs it is likely to resist, all from one test.

What Microarrays Cannot See

For all their power, microarrays have blind spots. You can only find what you put probes for, so novel or unexpected sequences that were not included in the array design will be invisible. In clinical cytogenetics, truly balanced chromosomal rearrangements (where DNA is rearranged but nothing is gained or lost) and very low-level mosaicism (where only a small fraction of cells carry an abnormality) are generally not detectable by arrays, though these account for fewer than 1% of abnormal findings in patients with developmental disabilities.¹¹

Microarrays also struggle with dynamic range. Because fluorescent signals saturate at high intensities, extremely abundant transcripts can max out the detector while low-abundance transcripts barely register above background noise. RNA sequencing (RNA-Seq) dramatically outperforms arrays on this front: one comparison found RNA-Seq delivered a dynamic range spanning from roughly 577-fold upregulation to 884-fold downregulation, while microarrays topped out at about 41-fold in either direction.²³ RNA-Seq can also detect entirely new transcripts and splice variants that no array was designed to look for.²⁴

The cost gap has narrowed but has not disappeared. Estimates have placed microarray costs at roughly $300 per sample versus up to $1,000 for RNA-Seq.²⁴ For studies that need to profile hundreds or thousands of samples and are looking at well-characterized gene sets rather than hunting for surprises, microarrays remain practical and cost-effective. For discovery work or when you need the widest possible view, sequencing has largely taken over.

The False Discovery Problem in Large-Scale Analysis

When you test thousands of genes at once, the mathematics of chance becomes a real concern. If you run standard statistical tests on 20,000 genes, you should expect roughly 1,000 false positives at a conventional significance threshold, even if nothing biologically interesting is happening. Microarray analysis highlighted this problem early and drove the widespread adoption of false discovery rate (FDR) correction methods in genomics.²⁵

Small experiments are especially vulnerable. With only five arrays per group and a high proportion of unchanged genes, the expected false discovery rate can climb above 60%, and it approaches 95% in the worst-case scenario.²⁶ This is why replication and adequate sample sizes matter so much in microarray experiments, and why published gene lists from underpowered studies should be treated with caution.

On the processing side, raw fluorescence data needs substantial computational cleanup before it can be interpreted. Background correction, normalization across arrays, and summarization of multiple probe signals into a single gene-level measurement are all standard steps. The robust multi-array average (RMA) method, which applies background adjustment, normalization, and log transformation, became one of the most widely used preprocessing pipelines for oligonucleotide arrays.²⁷

The Challenge of Uncertain Results

One of the thorniest issues with clinical microarray testing, especially in prenatal settings, is the discovery of variants of uncertain significance (VUS). These are copy number changes that the lab can see but cannot confidently classify as harmful or benign, because the scientific community simply does not have enough data yet. Reporting practices for these uncertain results vary between countries and even between hospitals within the same country.²⁸

For parents, receiving a VUS result during pregnancy can be deeply distressing. Research with families who received microarray results found that incomplete comprehension was a prominent issue regardless of whether the finding was classified as pathogenic or uncertain. Parents struggled with the scientific ambiguity and with what the result meant for their child’s future.²⁹ Ethicists have debated whether uncertain results should even be disclosed, but the prevailing view favors transparency: withholding information material to a patient’s decision-making would be paternalistic, and non-directive genetic counseling before and after testing is considered essential to help families navigate the uncertainty.³⁰

Quality Standards and Regulation

Clinical microarray platforms used for patient diagnosis are not unregulated. In the United States, the American College of Medical Genetics and Genomics (ACMG) publishes detailed technical standards for chromosomal microarray analysis, covering both constitutional (inherited) and neoplastic (cancer-related) applications. These were most recently updated in 2021.³¹ All platforms intended for clinical testing must either receive FDA clearance and be verified by the lab, or be validated internally if they are laboratory-developed tests. Validation requires documenting that the test performs as expected and achieves its intended result.³²

The FDA has its own perspective on microarray-based diagnostics, recognizing that translating these devices from research into clinical use raises unique challenges around ensuring safety and effectiveness given the sheer number of simultaneous measurements involved.³³

Beyond DNA and RNA

The microarray concept is not limited to nucleic acids. Protein microarrays pin down thousands of proteins on a chip and probe them for interactions, enzymatic activity, or antibody binding. These fall into two broad classes: analytical arrays, which capture and quantify proteins from a sample using known antibodies, and functional arrays, which display purified proteins and test what they do.³⁴

Antibody microarrays, a subtype of the analytical class, have become a popular tool for studying protein abundance, signaling pathways, and potential drug targets.³⁵ In autoimmune disease research, protein microarrays displaying up to 20,000 targets have been used to profile autoantibodies, helping identify antibodies directed against receptors and other proteins that play causative roles in disorders of the nervous system.³⁶ A third format, the reverse-phase protein array, flips the usual logic: instead of spotting capture molecules and applying the sample, it spots the cell or tissue lysate directly and then probes it with specific antibodies, which is useful for comparing a particular protein’s level across dozens of patient samples on a single slide.

Protein microarrays have not achieved the same commercial ubiquity as DNA arrays, partly because proteins are harder to manufacture, purify, and keep stable on a surface. But in specialized applications like autoimmune profiling and drug development screening, they fill a niche that nucleic acid methods cannot.