Box 1 | Array CGH versus SNP microarray detection

From the following article:

Genome structural variation discovery and genotyping

Can Alkan, Bradley P. Coe & Evan E. Eichler

Nature Reviews Genetics 12, 363-376 (May 2011)


In array comparative genomic hybridization (array CGH), the signal ratio between a test and reference sample is normalized and converted to a log2 ratio, which acts as a proxy for copy number18, 25, 112. An increased log2 ratio represents a gain in copy number in the test compared with the reference; conversely, a decrease indicates a loss in copy number (see the figure, part a). SNP arrays generate a similar metric by comparing the signal intensities for the sample being analysed to a collection of reference hybridizations, or the rest of the population being analysed (part b, upper panel). The log ratio metric for SNP arrays demonstrates a lower per-probe signal-to-noise ratio (SNR) than array CGH (compare a and b in the figure); however, SNP arrays offer an additional metric that enables a more comprehensive assignment of copy number than does array CGH. This metric, termed B allele frequency (BAF) (part b, lower panel), can be calculated as the proportion of the total allele signal (A + B) explained by a single allele (A). The BAF has a significantly higher per-probe SNR than the log ratio data and can be interpreted as follows: a BAF of 0 represents the genotype (A/A or A/–), whereas 0.5 represents (A/B) and 1 represents (B/B or B/–). Different BAF values occur for AAB and ABB genotypes or more complex genotypes (for example, AAAB, AABB and BBBA). Homozygous deletions result in a failure of the BAF to cluster23, 24. Thus, the BAF may be used to accurately assign copy numbers from 0 to 4 in diploid regions of the genome. The BAF also allows detection of copy-neutral events such as segmental uniparental disomy (segmental UPD) or whole-chromosome UPD and identity by descent (IBD), which results when a segment of one chromosome is replaced by the other allele without a change in copy number (this is therefore not detectable by array CGH)24. An additional advantage of the BAF is that it can be used to reliably detect and type low-level mosaic gains and losses24, 113, 114 (see the figure, part b).

Genome structural variation discovery and genotyping 

Another important consideration in choosing an array platform is the ability to detect alterations in the size range being investigated. Array resolution is complicated by non-uniform probe distributions and differing SNRs between platforms, and as a result two platforms cannot be compared by simply counting the number of probes included. The number of probes required to detect a single-copy alteration varies between platforms, with Agilent Technologies offering the highest per-probe performance25, 26, 32. Part c of the figure shows the probe coverage of several major array platforms as determined by ResCalc25. This represents the theoretical ability to detect a copy number variant at any given location in the genome. In practice, however, thresholds of copy-number detection are typically greater owing to variable probe performance (Box 2). Although alterations can, theoretically, be detected with a single probe using the Agilent platform, we set the detection limit to a more realistic (in a discovery context) three probes. The other major array platforms tend to require more probes, with Roche NimbleGen34 and Illumina16 platforms requiring ten probes, and Affymetrix39 requiring 20 probes to reliably detect a single-copy alteration.