PhyloDetect: a likelihood-based strategy for detecting microorganisms with diagnostic microarrays

Abstract

MOTIVATION: Detection and identification of microbes using diagnostic arrays is still subject of ongoing research. Existing significance-based algorithms consider an organism detected even if a significant number of the microarray probes that match the organism are called absent in a hybridization. Further, they do generate redundant results if the target organisms show high sequence similarity and the microarray probes cannot discriminate all of them. RESULTS: We propose a new analysis strategy that considers organism similarities and calls organisms only present if the probes that match the organism but are absent in a hybridization can be explained by random events. In our strategy, we.rst identify the groups of target organisms that are actually distinguishable by the array. Subsequently, these organism groups are placed in a hierarchical tree such that groups matching only less specific probes are closer to the tree root, and groups that are discriminated only by few probes are close to each other. Finally, we compute for each group a likelihood score that is based on a hypothesis test with the null hypothesis that the group was actually present in the hybridized sample. We have validated our strategy using datasets from two different array types and implemented it as an easy-to-use web application. AVAILABILITY: http://www.fgcz.ethz.ch/PhyloDetect. SUPPLEMENTARY INFORMATION: Example data is available at http://www.fgcz.ethz.ch/PhyloDetect.