With the recent and continued growth of publicly available sequence databases, diagnostic applications of sequence-based techniques such as PCR and DNA microarrays are currently of widespread interest. Efforts are underway to greatly expand global surveillance and pandemic preparedness for influenza, driving the need for rapid, low-cost analytical techniques capable of discriminating between several subtypes of the virus. Low-density oligonucleotide microarrays provide several desirable characteristics, but must be designed carefully to ensure that the limited probe set can detect the widest possible range of viruses in addition to discriminating between virus subtypes.

A new microarray probe design protocol was developed to specifically address large, highly variable sequence databases, such as those for influenza genes, using a reductionist approach. Databases were sorted into phylogenetic clusters containing similar viruses, then analyzed to find highly conserved regions of sequence within each cluster. Probe oligos were designed from these conserved regions. Several new pieces of software were written to aid in this design process, most notably ConFind, a tool for identifying conserved regions of sequence data from sequence alignments with missing or ambiguous sequence data. FluChip-55, designed using this methodology, allows three common subtypes of influenza to be distinguished. Further enhancements to the design protocol were developed to aid in developing several diagnostic microarrays, including a microarray for detecting antiviral resistance mutations and the MChip, an influenza subtyping microarray based on pattern recognition.

Comparisons of influenza M gene sequences and observed microarray intensity data revealed that the MChip probe oligos do not follow the trends observed for hybridizations in solution-like conditions in cases where the oligo does not perfectly match the influenza gene target. A new position-weighting model for mismatched hybridizations on microarray surfaces was proposed to better model the observed trends. In combination with a method for normalizing single-color microarray fluorescence intensities, the position-weighted mismatch model accurately predicted intensities observed for the MChip oligos by considering only the positions of mismatches between an oligo and the corresponding influenza M gene target.