Search

The application of multiple omics technologies in biomedical cohorts has the potential to reveal patient-level disease characteristics and individualized response to treatment. However, the scale and heterogeneous nature of multi-modal data makes integration and inference a non-trivial task. We developed a deep-learning-based framework, multi-omics variational autoencoders (MOVE), to integrate such data and applied it to a cohort of 789 people with newly diagnosed type 2 diabetes with deep multi-omics phenotyping from the DIRECT consortium. Using in silico perturbations, we identified drug–omics associations across the multi-modal datasets for the 20 most prevalent drugs given to people with type 2 diabetes with substantially higher sensitivity than univariate statistical tests. From these, we among others, identified novel associations between metformin and the gut microbiota as well as opposite molecular responses for the two statins, simvastatin and atorvastatin. We used the associations to quantify drug–drug similarities, assess the degree of polypharmacy and conclude that drug effects are distributed across the multi-omics modalities.

Background & Aims: Excess liver iron content is common and is linked to the risk of hepatic and extrahepatic diseases. We aimed to identify genetic variants influencing liver iron content and use genetics to understand its link to other traits and diseases. Methods: First, we performed a genome-wide association study (GWAS) in 8,289 individuals from UK Biobank, whose liver iron level had been quantified by magnetic resonance imaging, before validating our findings in an independent cohort (n = 1,513 from IMI DIRECT). Second, we used Mendelian randomisation to test the causal effects of 25 predominantly metabolic traits on liver iron content. Third, we tested phenome-wide associations between liver iron variants and 770 traits and disease outcomes. Results: We identified 3 independent genetic variants (rs1800562 [C282Y] and rs1799945 [H63D] in HFE and rs855791 [V736A] in TMPRSS6) associated with liver iron content that reached the GWAS significance threshold (p

Prediction of precipitation using simulations on various climate variables provided by Global Climate Models (GCM) as covariates is often required for regional hydrological assessment studies. In this paper, we use a sufficient dimension reduction method to analyze monthly precipitation data over the Missouri River Basin (MRB). At each location, effective reduced sets of monthly historical simulated data from a neighborhood provided by MIROC5, a Global Climate Model, are first obtained via a semi-continuous adaptation of the Sliced Inverse Regression, a sufficient dimension reduction approach. These reduced sets are used subsequently in a modified Nadaraya-Watson method for prediction. We implement the method on a computing cluster, and demonstrate that it is scalable. We observe a signficant speedup in the runtime when implemented in parallel. This is an attractive alternative to the traditional spatio-temporal analysis of the entire region given the large number of locations and temporal instances.

Alzheimer's Disease is a neurological disorder chiefly present in the elderly that affects functions of the brain such as memory and logic, eventually resulting in death. There is no known cure to Alzheimer's and evidence points to the possibility of a genetic link. This study analyzes microarray data from patients with Alzheimer's disease and disease-free patients in order to evaluate and determine differential gene expression patterns between the two groups. The statistical problem stemming from this data involves many predictor variables with a small sample size, preventing the use of classical statistical approaches from being effective. We turn to a novel three-step approach: first, we screen the genes in order to keep only the genes marginally related to the outcome (presence of Alzheimer's); second, we implemented a sparse sufficient dimension reduction to retain only predictors relevant to the outcome; lastly, we perform a hierarchical clustering method to group genes that exhibit mutual dependence. We adapted this methodology from Adragni et. al and expand on their work by optimizing the existing R code with parallel capabilities in order to enhance performance speed. Thus, our results reflect both an analysis of the microarray data and a performance study of the modified code.

The second most expensive part of the 2010 Decennial Census was Address Canvassing (AdCan), a eld operation to prepare the Master Address File (MAF) for census day. The MAF is a database of households in the United States maintained by the US Census Bureau and is used as a basis for the census and household surveys that it conducts. Motivated by the importance of the MAF and the cost of a large scale AdCan operation, there is an interest to use statistical methodologies to explain MAF errors discovered during canvassing. Ideally, statistical models could be used to predict future errors and assist with updating of the MAF. A major challenge in constructing a MAF error model is that important predictor variables associated with MAF errors are not known. Some recent works at Census Bureau have carried out variable selection using a collection of data sources, treating counts of errors per census block as the outcome. It may be possible to use dimension reduction methodologies to obtain count models with much lower dimensional predictors. Adragni et al. [4] proposed a methodology called Minimum Average Deviance Estimation (MADE), which is based on the concept of local regression and embeds sufficient dimension reduction of the predictors. MADE assumes a forward regression with the response variable following an exponential family distribution, such as Poisson for counts. The goal of this project is to evaluate the performance of MADE on large data sets using simulations. We parallelized several snippets of the MADE source code to improve its performance and compare the speed up of these parallelized snippets with their serial alternatives. Simulated data sets with increasing dimensions are used to evaluate the run time. A limited stress test is performed to determine the extent of problem size that MADE can handle on maya, a high performance computing cluster at UMBC. These tests allow us to evaluate the capabilities of MADE to scale to large data sets, such as the AdCan modeling problem.

The study of allostery, a regulatory process that occurs in complex macromolecules such as proteins, is of particular interest as it has a key role in determining the function of these macromolecules. Allostery produces motional correlations that can be analyzed using different statistical methods. We implement a program in the statistical programming language R that uses polynomial regression and leave-one-out cross-validation to model relationships in data obtained from different sites in the protein, using the square root of the coefficient of determination to detect both linear and non-linear trends. The performance of the program will be studied on a simulated data set with linear and non-linear relationships and the effectiveness of the implemented methods as it relates to this problem will be assessed.

Complex biomolecules such as proteins can respond to changes in their environment through a process called allostery, which plays an important role in regulating the function of these biomolecules. Allostery occurs when an event at a specific location in a macromolecule produces an effect at a location in the molecule some distance away. An important component of allostery is the coupling of protein sites. Such coupling is one mechanism by which allosteric effects can be transmitted over long distances. To understand this phenomenon, molecular dynamic simulations are carried out with a large number of atoms, and the trajectories of these atoms are recorded over time. Simple correlation methods have been used in the literature to identify coupled motions between protein sites. We implement a recently developed statistical method for dimension reduction called principal fitted components (PFC) in the statistical programming language R to identify both linear and non-linear correlations between protein sites while dealing efficiently with the high dimensionality of the data. PFC models reduce the dimensionality of data while capturing linear and nonlinear dependencies among predictors (atoms) using a flexible set of basis functions. For faster processing, we implement the PFC algorithm using parallel computing through the Programming with Big Data in R (pbdR) package for R. We demonstrate the methods’ effectiveness on simulated datasets, and apply the routine to time series data from Molecular Dynamic (MD) simulations to identify coupled motion among the atoms.

Programming with big data in R (pbdR), a package used to implement high-performance computing in the statistical software R, uses block cyclic distribution to organize large data across many processes. Because computations performed on large matrices are often not associative, a systematic approach must be used during parallelization to divide the matrix correctly. The block cyclic distribution method stresses a balanced load across processes by allocating sections of data to a corresponding node. This method achieves well divided data that each process computes individually and calculates a final result more efficiently. A nontrivial problem occurs when using block cyclic distribution: Which combinations of different block sizes and grid layouts are most effective? These two factors greatly influence computational efficiency, and therefore it is crucial to study and understand their relationship. To analyze the effects of block size and processor grid layout, we carry out a performance study of the block cyclic process used to compute a principal components analysis (PCA). We apply PCA both to a large simulated data set and to data involving the analysis of single nucleotide polymorphisms (SNPs). We implement analysis of variance (ANOVA) techniques in order to distinguish the variability associated with each grid layout and block distribution. Once the nature of these factors is determined, predictions about the performance for much larger data sets can be made. Our final results demonstrate the relationship between computational efficiency and both block distribution and processor grid layout, and establish a benchmark regarding which combinations of these factors are most effective.

The Missouri River Basin (MRB) is an important food-producing region in the United States and Canada. Climate variability and water availability affect crops production in this region. Past climate data have been recorded at various locations in the basin over a period of ten years. We use the data for a retrospective prediction of rainfall. As the dimension of the data is relatively large, a sufficient dimension reduction approach is used to reduce the dimensionality of the data while preserving the regression information pertinent to rainfall. We use the nascent dimension reduction methodology called Minimum Average Deviance Estimation or MADE to reduce the dimensionality of the climate data. Since MADE is still a tool in development, we explored two of its intrinsic prediction methods and compared them to the Nadaraya-Watson prediction approach by a cross-validation. A parallel implementation of MADE and its prediction methods on a high performance computer were carried out. A performance study was performed along with the application of the best prediction method to the MRB climate data.

The U.S. Geological Survey National Water-Quality Assessment Project conducted a study of 225 sites in the Chesapeake Bay watershed to estimate base flow. Baseflow is the estimated volumetric discharge of water, primarily from groundwater sources, that is relayed to the measurement sites. The study is necessary in order to address the nation’s water supply for changes in the environment. Baseflow is estimated using a recursive digital filter. Calculating the variability of baseflow water discharge is important to make informed decisions about water regulation. We explored the estimation of variability of baseflow using two methods: the bootstrap method and the Delta method. Each method has its own limitations and requirements. Ultimately, bootstrapping was shown to be a reasonable recommendation for estimating baseflow variability. The bootstrapping algorithm was parallelized in order to compute numerous iterations on multiple processors for big data analysis. The derivation of the variability of a non-constant streamflow was also considered for further study, but not implemented.