MOTIVATION: Current methods in diagnostic microbiology typically focus on the detection of a single genomic locus or protein in a candidate agent. The presence of the entire microbe is then inferred from this isolated result. Problematically, the presence of recombination in microbial genomes would go undetected unless other genomic loci or protein components were specifically assayed. Microarrays lend themselves well to the detection of multiple loci from a given microbe; furthermore, the inherent nature of microarrays facilitates highly parallel interrogation of multiple microbes. However, none of the existing methods for analyzing diagnostic microarray data has the capacity to specifically identify recombinant microbes. In previous work, we developed a novel algorithm, VIPR, for analyzing diagnostic microarray data. RESULTS: We have expanded upon our previous implementation of VIPR by incorporating a hidden Markov model (HMM) to detect recombinant genomes. We trained our HMM on a set of non-recombinant parental viruses and applied our method to 11 recombinant alphaviruses and 4 recombinant flaviviruses hybridized to a diagnostic microarray in order to evaluate performance of the HMM. VIPR HMM correctly identified 95% of the 62 inter-species recombination breakpoints in the validation set and only two false-positive breakpoints were predicted. This study represents the first description and validation of an algorithm capable of detecting recombinant viruses based on diagnostic microarray hybridization patterns. AVAILABILITY: VIPR HMM is freely available for academic use and can be downloaded from http://ibridgenetwork.org/wustl/vipr. CONTACT: [email protected]. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Determining whether a user or system is exercising appropriate security practices is difficult in any context. Such difficulties are particularly pronounced when uncontrolled or unknown platforms join public networks. Commonly practiced techniques used to vet these hosts, such as system scans, have the potential to infringe on the privacy of users. In this article, we show that it is possible for clients to prove both the presence and proper functioning of security infrastructure without allowing unrestricted access to their system. We demonstrate this approach, specifically applied to antivirus security, by requiring clients seeking admission to a network to positively identify the presence or absence of malcode in a series of puzzles. The implementation of this mechanism and its application to real networks are also explored. In so doing, we demonstrate that it is not necessary for an administrator to be invasive to determine whether a client implements required security practices.

Slow learning of neural-network function approximators can frequently be attributed to interference, which occurs when learning in one area of the input space causes unlearning in another area. To mitigate the effect of unlearning, this paper develops an algorithm that adjusts the weights of an arbitrary, nonlinearly parameterized network such that the potential for future interference during learning is reduced. This is accomplished by the reduction of a biobjective cost function that combines the approximation error and a term that measures interference. An analysis of the algorithm's convergence properties shows that learning with this algorithm reduces future unlearning. The algorithm can be used either during online learning or can be used to condition a network to have immunity from interference during a future learning stage. A simple example demonstrates how interference manifests itself in a network and how less interference can lead to more efficient learning. Simulations demonstrate how this new learning algorithm speeds up the training in various situations due to the extra cost function term.

Interference in neural networks occurs when learning in one area of the input space causes unlearning in another area. Networks that are less susceptible to interference are referred to as spatially local networks. To obtain a better understanding of these properties, a theoretical framework, consisting of a measure of interference and a measure of network localization, is developed. These measures incorporate not only the network weights and architecture but also the learning algorithm. Using this framework to analyze sigmoidal, multilayer perceptron (MLP) networks that employ the backpropagation learning algorithm on the quadratic cost function, we address a familiar misconception that single-hidden-layer sigmoidal networks are inherently nonlocal by demonstrating that given a sufficiently large number of adjustable weights, single-hidden-layer sigmoidal MLP's exist that are arbitrarily local and retain the ability to approximate any continuous function on a compact domain.