Improved Probe Selection for DNA Arrays Using Nonparametric Kernel Density Estimation

Oligonucleotide fingerprinting of rRNA genes (OFRG) is a method for identifying arrayed ribosomal RNA genes (rDNA) through a series of hybridization experiments with short oligonucleotide probes. Due to its low cost and high speed, it is an effective tool for analyzing microbial communities. OFRG relies on probe sets that can discriminate large collections of clones. Although the currently used probe design algorithm produces probe sets whose theoretical accuracy is close to optimum, those probes often do not hybridize in a consistent and predictable manner in actual biological experiments. We assume that these failures occur following an unknown probability distribution. In this paper, a nonparametric kernel density estimation method is proposed to estimate this distribution and to predict probe reliability. These predictions are used to reduce the number of unreliable probes chosen by the probe design algorithms. Our preliminary results show that the application of this method leads to a significant decrease in the number of unreliable probes