Bacterial multistrain genomotyping arrays: optimal choice of control in two-colour experiments

Abstract number: O366

Pinto F., Aguiar S., Melo-Cristino J., Ramirez M.

Objectives: Comparative genomic hybridisation (aCGH) studies use microarrays to evaluate the distribution of genes of sequenced bacterial strains among unsequenced strains. The use of this technology to study different strains of the same species has led to important insights into the identification of the species "core-genome" as well as the distribution of the genes in the "accessory genome" in the population. As genomic sequences from multiple strains of the same species become available, multistrain microarrays are designed, containing spots for every unique gene in the sequenced strains. In two-colour aCGH experiments with multistrain microarrays, the choice of control sample can be the genomic DNA of one strain or a mix of the genomic DNA of all the strains used in the array design. This important problem has no universally accepted solution. The aim of this study was to evaluate the relative performance of each of the two types of control in two-colour aCGH experiments.

Methods: We performed a comparative study of the two control sample options with a microarray designed with three fully sequenced strains. We hybridised two of these strains as test samples using only the third strain as a control or a mix of the three strains as the control sample. The strain used in the single control was the one with more specific spots in the array. Resulting Log-ratios of test over control sample signal were used to classify genes as present or absent. We evaluated classification performance through the area under a receiver operating characteristic curve (AUC), accuracy, sensitivity and specificity. AUC measures the probability that the Log-ratio of a present gene is higher than that of an absent gene.

Results: We show that for both types of control sample it is beneficial to analyse spots in separate classes according to their expected control channel signal (0.05–0.15 AUC increase). The use of a mix control leads to higher accuracies (0.05 increase). This best performance is due to gains in sensitivity (0.21 increase, p = 0.001) that compensate minor losses in specificity (0.05 decrease, p = 0.014). Moreover, the use of a single strain control increases the error rate in genes that are not present in all reference strains, the set of genes where more variation across unsequenced strains is expected.

Conclusion: The use of a mix control in multistrain microarrays leads to better performances than using a single strain control.

Session Details

Date: 19/04/2008
Time: 00:00-00:00
Session name: 18th European Congress of Clinical Microbiology and Infectious Diseases
Location: Barcelona, Spain
Presentation type:
Back to top