We propose several means for improving the performance and training of neural networks for classification. We use crossvalidation as a tool for optimizing network parameters and architecture. We show further that the remaining residual "generalization" error can be reduced by invoking ensembles of similar networks.Zndex Terms-Crossvalidation, fault tolerant computing, neural networks, N-version programming.

Viruses are the most common biological entities in the marine environment. There has not been a global survey of these viruses, and consequently, it is not known what types of viruses are in Earth's oceans or how they are distributed. Metagenomic analyses of 184 viral assemblages collected over a decade and representing 68 sites in four major oceanic regions showed that most of the viral sequences were not similar to those in the current databases. There was a distinct “marine-ness” quality to the viral assemblages. Global diversity was very high, presumably several hundred thousand of species, and regional richness varied on a North-South latitudinal gradient. The marine regions had different assemblages of viruses. Cyanophages and a newly discovered clade of single-stranded DNA phages dominated the Sargasso Sea sample, whereas prophage-like sequences were most common in the Arctic. However most viral species were found to be widespread. With a majority of shared species between oceanic regions, most of the differences between viral assemblages seemed to be explained by variation in the occurrence of the most common viral species and not by exclusion of different viral genomes. These results support the idea that viruses are widely dispersed and that local environmental conditions enrich for certain viral types through selective pressure.

Viruses are the most common biological entities in the oceans by an order of magnitude. However, very little is known about their diversity. Here we report a genomic analysis of two uncultured marine viral communities. Over 65% of the sequences were not significantly similar to previously reported sequences, suggesting that much of the diversity is previously uncharacterized. The most common significant hits among the known sequences were to viruses. The viral hits included sequences from all of the major families of dsDNA tailed phages, as well as some algal viruses. Several independent mathematical models based on the observed number of contigs predicted that the most abundant viral genome comprised 2-3% of the total population in both communities, which was estimated to contain between 374 and 7,114 viral types. Overall, diversity of the viral communities was extremely high. The results also showed that it would be possible to sequence the entire genome of an uncultured marine viral community.

Mucosal surfaces are a main entry point for pathogens and the principal sites of defense against infection. Both bacteria and phage are associated with this mucus. Here we show that phageto-bacteria ratios were increased, relative to the adjacent environment, on all mucosal surfaces sampled, ranging from cnidarians to humans. In vitro studies of tissue culture cells with and without surface mucus demonstrated that this increase in phage abundance is mucus dependent and protects the underlying epithelium from bacterial infection. Enrichment of phage in mucus occurs via binding interactions between mucin glycoproteins and Ig-like protein domains exposed on phage capsids. In particular, phage Ig-like domains bind variable glycan residues that coat the mucin glycoprotein component of mucus. Metagenomic analysis found these Ig-like proteins present in the phages sampled from many environments, particularly from locations adjacent to mucosal surfaces. Based on these observations, we present the bacteriophage adherence to mucus model that provides a ubiquitous, but non-host-derived, immunity applicable to mucosal surfaces. The model suggests that metazoan mucosal surfaces and phage coevolve to maintain phage adherence. This benefits the metazoan host by limiting mucosal bacteria, and benefits the phage through more frequent interactions with bacterial hosts. The relationships shown here suggest a symbiotic relationship between phage and metazoan hosts that provides a previously unrecognized antimicrobial defense that actively protects mucosal surfaces.symbiosis | host-pathogen | virus | immunoglobulin | immune system

Here we present the first metagenomic analyses of an uncultured viral community from human feces, using partial shotgun sequencing. Most of the sequences were unrelated to anything previously reported. The recognizable viruses were mostly siphophages, and the community contained an estimated 1,200 viral genotypes.

Microbial viruses can control host abundances via density-dependent lytic predator-prey dynamics. Less clear is how temperate viruses, which coexist and replicate with their host, influence microbial communities. Here we show that virus-like particles are relatively less abundant at high host densities. This suggests suppressed lysis where established models predict lytic dynamics are favoured. Meta-analysis of published viral and microbial densities showed that this trend was widespread in diverse ecosystems ranging from soil to freshwater to human lungs. Experimental manipulations showed viral densities more consistent with temperate than lytic life cycles at increasing microbial abundance. An analysis of 24 coral reef viromes showed a relative increase in the abundance of hallmark genes encoded by temperate viruses with increased microbial abundance. Based on these four lines of evidence, we propose the Piggyback-the-Winner model wherein temperate dynamics become increasingly important in ecosystems with high microbial densities; thus 'more microbes, fewer viruses'.

New expressions for the availability dissipated in a finite-time endoreversible process are found by use of Weinhold's metric on ecluilibrium states of a thermodynamic system. In particular, the dissipated availability is given by the square of the length of the corresponding curve, times a mean relaxation time, divided by the total time of the process. '&he results extend to local thermodynamic equilibrium if instead of length one uses distance (length of the shortest curve) between initial and final states. PACS numbers: 05.70.-aThe results presented below give an important tool for finding limits on the efficiency of' operation of thermodynamic processes in finite time.We have been pursuing various approaches to this problem for several years, ' but the results below give a new inroad for a remarkably general class of processes by providing expressions for the inherent irreversibility quantified by the loss of available work: the availability not transformed into work during the process. This dissiPated availabil. ity is sometimes called "irreversibility. '" The dissipated availability dA" is related to the entropy production dS"by dA"= TdS". We derive expressions for the availability dissipated when a thermodynamic system undergoes a process during which it may be assumed to be in internal equilibrium, though interacting with an environment, which is also in equilibrium. Since the dissipated availability is an extensive quantity, an extension of our expressions to local thermodynamic equilibrium is immediate. These expressions involve the thermodynamic length introduced by Weinhold" and hint at the existence of a temporal element in the classical formalism dealing only with equilibrium.We assume that the time scales for internal relaxation of system and surroundings are much shorter than the time scale on which system and surroundings interact. This implies that we may consider both system and surroundings to be in equilibrium states at each instant of time. This assumption is already among the postulates for local thermodynamic equilibrium. The macroscopic form of this assumption was introduced by Rubin'. a process is e~do~evexsible provided the subsystems participating in the process are in internal equilibrium at each instant. The expression given below for dissipation in a shock wave hints that our expressions for the dissipated availability are valid in a context wider than our derivations based on endoreversibility show. Besides the total time for the process, our expressions for the dissipated availability include only one nonequilibrium parameter: the mean relaxation time. The interpretation of this relaxation time is straightforward for processes during which the system and its environment are close to equilibrium with each other. This automatically holds (except perhaps at the boundary) if the local thermodynamic equilibrium model is appropriate.We now define the notion of thermodynamic length. The second-derivative matrix of the internal energy U with respect to extensive variables X =(X".. . , X")...

Recent studies have highlighted the surprising richness of soil bacterial communities; however, bacteria are not the only microorganisms found in soil. To our knowledge, no study has compared the diversities of the four major microbial taxa, i.e., bacteria, archaea, fungi, and viruses, from an individual soil sample. We used metagenomic and small-subunit RNA-based sequence analysis techniques to compare the estimated richness and evenness of these groups in prairie, desert, and rainforest soils. By grouping sequences at the 97% sequence similarity level (an operational taxonomic unit [OTU]), we found that the archaeal and fungal communities were consistently less even than the bacterial communities. Although total richness levels are difficult to estimate with a high degree of certainty, the estimated number of unique archaeal or fungal OTUs appears to rival or exceed the number of unique bacterial OTUs in each of the collected soils. In this first study to comprehensively survey viral communities using a metagenomic approach, we found that soil viruses are taxonomically diverse and distinct from the communities of viruses found in other environments that have been surveyed using a similar approach. Within each of the four microbial groups, we observed minimal taxonomic overlap between sites, suggesting that soil archaea, bacteria, fungi, and viruses are globally as well as locally diverse.