Discriminatory power, typability, and accuracy of single nucleotide extension microarrays

Despite great conceptual promise, the use of microarrays in typing approaches has not yet gained wide acceptance. The establishment of proper criteria for determining discriminatory power as well as typability and the accuracy of microarray data remains to be solved. Purely experimental estimations of these parameters would far exceed what is experimentally practical. We therefore used simulations in combination with experimental results in parameter estimations. Our assay was based on 26 single nucleotide polymorphisms (SNPs) identified in the Campylobacterjejuni Multi Locus Sequence Typing (MLST) database (http:///pubmist.org/campylobacter/). The SNPs were detected using a single nucleotide extension (SNE) typing microarray. Unknown isolates were assigned to the known sequence type(s) by calculating weighted sum of matches minus a weighted sum of mismatches between predicted and candidate genotype. The weights were set according to the Bayesian posterior probability of the SNP classification. These studies showed that any typing or profiling method based on binary data requires an accuracy of <2-3% error for each datapoint (in our case SNPs) to classify the isolates to the correct allelic profile in 90% of the cases. The classification error for our experimental data was 3.2% (after removing 5 high error SNPs). We therefore conclude that SNE microarrays are promising for future high-throughput typing of bacteria.