Host gene expression classifiers diagnose acute respiratory illness etiology.

Acute respiratory infections caused by bacterial or viral pathogens are among the most common reasons for seeking medical care. Despite improvements in pathogen-based diagnostics, most patients receive inappropriate antibiotics. Host response biomarkers offer an alternative diagnostic approach to direct antimicrobial use. This observational cohort study determined whether host gene expression patterns discriminate noninfectious from infectious illness and bacterial from viral causes of acute respiratory infection in the acute care setting. Peripheral whole blood gene expression from 273 subjects with community-onset acute respiratory infection (ARI) or noninfectious illness, as well as 44 healthy controls, was measured using microarrays. Sparse logistic regression was used to develop classifiers for bacterial ARI (71 probes), viral ARI (33 probes), or a noninfectious cause of illness (26 probes). Overall accuracy was 87% (238 of 273 concordant with clinical adjudication), which was more accurate than procalcitonin (78%, P < 0.03) and three published classifiers of bacterial versus viral infection (78 to 83%). The classifiers developed here externally validated in five publicly available data sets (AUC, 0.90 to 0.99). A sixth publicly available data set included 25 patients with co-identification of bacterial and viral pathogens. Applying the ARI classifiers defined four distinct groups: a host response to bacterial ARI, viral ARI, coinfection, and neither a bacterial nor a viral response. These findings create an opportunity to develop and use host gene expression classifiers as diagnostic platforms to combat inappropriate antibiotic use and emerging antibiotic resistance.

Bayesian Learning with Dependency Structures via Latent Factors, Mixtures, and Copulas

Bayesian methods offer a flexible and convenient probabilistic learning framework to extract interpretable knowledge from complex and structured data. Such methods can characterize dependencies among multiple levels of hidden variables and share statistical strength across heterogeneous sources. In the first part of this dissertation, we develop two dependent variational inference methods for full posterior approximation in non-conjugate Bayesian models through hierarchical mixture- and copula-based variational proposals, respectively. The proposed methods move beyond the widely used factorized approximation to the posterior and provide generic applicability to a broad class of probabilistic models with minimal model-specific derivations. In the second part of this dissertation, we design probabilistic graphical models to accommodate multimodal data, describe dynamical behaviors and account for task heterogeneity. In particular, the sparse latent factor model is able to reveal common low-dimensional structures from high-dimensional data. We demonstrate the effectiveness of the proposed statistical learning methods on both synthetic and real-world data.

Distributed Feature Selection in Large n and Large p Regression Problems

Fitting statistical models is computationally challenging when the sample size or the dimension of the dataset is huge. An attractive approach for down-scaling the problem size is to first partition the dataset into subsets and then fit using distributed algorithms. The dataset can be partitioned either horizontally (in the sample space) or vertically (in the feature space), and the challenge arise in defining an algorithm with low communication, theoretical guarantees and excellent practical performance in general settings. For sample space partitioning, I propose a MEdian Selection Subset AGgregation Estimator ({\em message}) algorithm for solving these issues. The algorithm applies feature selection in parallel for each subset using regularized regression or Bayesian variable selection method, calculates the `median' feature inclusion index, estimates coefficients for the selected features in parallel for each subset, and then averages these estimates. The algorithm is simple, involves very minimal communication, scales efficiently in sample size, and has theoretical guarantees. I provide extensive experiments to show excellent performance in feature selection, estimation, prediction, and computation time relative to usual competitors.

While sample space partitioning is useful in handling datasets with large sample size, feature space partitioning is more effective when the data dimension is high. Existing methods for partitioning features, however, are either vulnerable to high correlations or inefficient in reducing the model dimension. In the thesis, I propose a new embarrassingly parallel framework named {\em DECO} for distributed variable selection and parameter estimation. In {\em DECO}, variables are first partitioned and allocated to m distributed workers. The decorrelated subset data within each worker are then fitted via any algorithm designed for high-dimensional problems. We show that by incorporating the decorrelation step, DECO can achieve consistent variable selection and parameter estimation on each subset with (almost) no assumptions. In addition, the convergence rate is nearly minimax optimal for both sparse and weakly sparse models and does NOT depend on the partition number m. Extensive numerical experiments are provided to illustrate the performance of the new framework.

For datasets with both large sample sizes and high dimensionality, I propose a new "divided-and-conquer" framework {\em DEME} (DECO-message) by leveraging both the {\em DECO} and the {\em message} algorithm. The new framework first partitions the dataset in the sample space into row cubes using {\em message} and then partition the feature space of the cubes using {\em DECO}. This procedure is equivalent to partitioning the original data matrix into multiple small blocks, each with a feasible size that can be stored and fitted in a computer in parallel. The results are then synthezied via the {\em DECO} and {\em message} algorithm in a reverse order to produce the final output. The whole framework is extremely scalable.

Floral and faunal characteristics and content of photosynthetic pigment in soils of northern Victoria Land, Antarctica

Although soil algae are among the main primary producers in most terrestrial ecosystems of continental Antarctica, there are very few quantitative studies on their relative proportion in the main algal groups and on how their distribution is affected by biotic and abiotic factors. Such knowledge is essential for understanding the functioning of Antarctic terrestrial ecosystems. We therefore analyzed biological soil crusts from northern Victoria Land to determine their pH, electrical conductivity (EC), water content (W), total and organic C (TC and TOC) and total N (TN) contents, and the presence and abundance of photosynthetic pigments. In particular, the latter were tested as proxies for biomass and coarse-resolution community structure. Soil samples were collected from five sites with known soil algal communities and the distribution of pigments was shown to reflect differences in the relative proportions of Chlorophyta, Cyanophyta and Bacillariophyta in these sites. Multivariate and univariate models strongly indicated that almost all soil variables (EC, W, TOC and TN) were important environmental correlates of pigment distribution. However, a significant amount of variation is independent of these soil variables and may be ascribed to local variability such as changes in microclimate at varying spatial and temporal scales. There are at least five possible sources of local variation: pigment preservation, temporal variations in water availability, temporal and spatial interactions among environmental and biological components, the local-scale patchiness of organism distribution, and biotic interactions.

(Table 1) N-isotope record of planktonc foraminifera of sediment core MD02-2550C2

Constraining variations in marine N2-fixation over glacial-interglacial timescales is crucial for determining the role of the marine nitrogen cycle in modifying ocean productivity and climate, yet paleo-records from N2-fixation regions are sparse. Here we present new nitrogen isotope (d15N) records of bulk sediment and foraminifera test-bound (FB) nitrogen extending back to the last ice age from the oligotrophic Gulf of Mexico (GOM). Previous studies indicate a substantial terrestrial input during the last ice age and early deglacial, for which we attempt to correct the bulk sediment d15N using its observed relationship with the C/N ratio. Both corrected bulk and FB-d15N reveal a substantial glacial-to-Holocene decrease of d15N toward Holocene values of around 2.5 per mil, similar to observations from the Caribbean. This d15N change is most likely due to a glacial-to-Holocene increase in regional N2-fixation. A deglacial peak in the FB-d15N of thermocline dwelling foraminifera Orbulina universa probably reflects a whole ocean increase in the d15N of nitrate during deglaciation. The d15N of the surface dwelling foraminifera Globigerinoides ruber and the corrected bulk d15N show little sign of this deglacial peak, both decreasing from last glacial values much earlier than does the d15N of O. universa; this may indicate that G. ruber and bulk N reflect the euphotic zone signal of an early local increase in N2-fixation. Our results add to the evidence that, during the last ice age, the larger iron input from dust did not lead to enhanced N2-fixation in this region. Rather, the glacial-to-Holocene decrease in d15N is best explained by a response of N2-fixation within the Atlantic to the deglacial increase in global ocean denitrification.

Seafloor spreading magnetic anomaly isochron map compilation for the Weddell Sea and Scotia Sea

The deep sea sedimentary record is an archive of the pre-glacial to glacial development of Antarctica and changes in climate, tectonics and ocean circulation. Identification of the pre-glacial, transitional and full glacial components in the sedimentary record is necessary for ice sheet reconstruction and to build circum-Antarctic sediment thickness grids for past topography and bathymetry reconstructions, which constrain paleoclimate models. A ~3300 km long Weddell Sea to Scotia Sea transect consisting of multichannel seismic reflection data from various organisations, were used to interpret new horizons to define the initial basin-wide seismostratigraphy and to identify the pre-glacial to glacial components. We mapped seven main units of which three are in the inferred Cretaceous-Paleocene pre-glacial regime, one in the Eocene-Oligocene transitional regime and three units in the Miocene-Pleistocene full glacial climate regime. Sparse borehole data from ODP leg 113 and SHALDRIL constrain the ages of the upper three units. Compiled seafloor spreading magnetic anomalies constrain the basement ages and the hypothetical age model. In many cases, the new horizons and stratigraphy contradict the interpretations in local studies. Each seismic sedimentary unit and its associated base horizon are continuous and traceable for the entire transect length, but reflect a lateral change in age whilst representing the same deposition process. The up to 1240 m thick pre-glacial seismic units form a mound in the central Weddell Sea basin and, in conjunction with the eroded flank geometry, support the interpretation of a Cretaceous proto-Weddell Gyre. The base reflector of the transitional seismic unit, which marks the initial ice sheet advances to the outer shelf, has a lateral model age of 26.6-15.5 Ma from southeast to northwest. The Pliocene-Pleistocene glacial deposits reveals lower sedimentations rates, indicating a reduced sediment supply. Sedimentation rates for the pre-glacial, transitional and full glacial components are highest around the Antarctic Peninsula, indicating higher erosion and sediment supply of a younger basement. We interpret an Eocene East Antarctic Ice Sheet expansion, Oligocene grounding of the West Antarctic Ice Sheet and Early Miocene grounding of the Antarctic Peninsula Ice Sheet.

Monthly air temperatures, snow density and snow and ice thickness at Malcolm Ramsay Lake, North America

In northern regions where observational data is sparse, lake ice models are ideal tools as they can provide valuable information on ice cover regimes. The Canadian Lake Ice Model was used to simulate ice cover for a lake near Churchill, Manitoba, Canada throughout the 2008/2009 and 2009/2010 ice covered seasons. To validate and improve the model results, in situ measurements of the ice cover through both seasons were obtained using an upward-looking sonar device Shallow Water Ice Profiler (SWIP) installed on the bottom of the lake. The SWIP identified the ice-on/off dates as well as collected ice thickness measurements. In addition, a digital camera was installed on shore to capture images of the ice cover through the seasons and field measurements were obtained of snow depth on the ice, and both the thickness of snow ice (if present) and total ice cover. Altering the amounts of snow cover on the ice surface to represent potential snow redistribution affected simulated freeze-up dates by a maximum of 22 days and break-up dates by a maximum of 12 days, highlighting the importance of accurately representing the snowpack for lake ice modelling. The late season ice thickness tended to be under estimated by the simulations with break-up occurring too early, however, the evolution of the ice cover was simulated to fall between the range of the full snow and no snow scenario, with the thickness being dependant on the amount of snow cover on the ice surface.

Radiocarbon age, Mg/Ca and d18O measurements on planktonic foraminifera of sediment core GeoB12605-3

The sea surface temperature (SST) of the tropical Indian Ocean is a major component of global climate teleconnections. While the Holocene SST history is documented for regions affected by the Indian and Arabian monsoons, data from the near-equatorial western Indian Ocean are sparse. Reconstructing past zonal and meridional SST gradients requires additional information on past temperatures from the western boundary current region. We present a unique record of Holocene SST and thermocline depth variations in the tropical western Indian Ocean as documented in foraminiferal Mg/Ca ratios and d18O from a sediment core off northern Tanzania. For Mg/Ca and thermocline d18O, most variance is concentrated in the centennial to bicentennial periodicity band. On the millennial time scale, an early to mid-Holocene (~7.8-5.6 ka) warm phase is followed by a temperature drop by up to 2°C, leading to a mid-Holocene cool interval (5.6-4.2 ka). The shift is accompanied by an initial reduction in the difference between surface and thermocline foraminiferal d18O, consistent with the thickening of the mixed layer and suggestions of a strengthened Walker circulation. However, we cannot confirm the expected enhanced zonal SST gradient, as the cooling of similar magnitude had previously been found in SSTs from the upwelling region off Sumatra and in Flores air temperatures. The SST pattern probably reflects the tropical Indian Ocean expression of a large-scale climate anomaly rather than a positive Indian Ocean Dipole-like mean state.