Filtering by
- Language: English
This dissertation focuses on developing methods for three different aspects of predictive analytics related to the disorder: automatic diagnosis, prognosis, and prediction of long-term treatment outcome. The data used for each task have their specific characteristics and demonstrate unique problems. Automatic diagnosis of melancholic depression is made on the basis of metabolic profiles and micro-array gene expression profiles where the presence of missing values and strong empirical correlation between the variables is not unusual. To deal with these problems, a method of generating a representative set of features is proposed. Prognosis is made on data collected from rating scales and questionnaires which consist mainly of categorical and ordinal variables and thus favor decision tree based predictive models. Decision tree models are known for the notorious problem of overfitting. A decision tree pruning method that overcomes the shortcomings of a greedy nature and reliance on heuristics inherent in traditional decision tree pruning approaches is proposed. The method is further extended to prune Gradient Boosting Decision Tree and tested on the task of prognosis of treatment outcome. Follow-up studies evaluating the long-term effect of the treatments on patients usually measure patients' depressive symptom severity monthly, resulting in the actual time of relapse upper bounded by the observed time of relapse. To resolve such uncertainty in response, a general loss function where the hypothesis could take different forms is proposed to predict the risk of relapse in situations where only an interval for time of relapse can be derived from the observed data.
In this dissertation, I carry out the research along the direction with particular focuses on scaling up the optimization of sparse learning for supervised and unsupervised learning problems. For the supervised learning, I firstly propose an asynchronous parallel solver to optimize the large-scale sparse learning model in a multithreading environment. Moreover, I propose a distributed framework to conduct the learning process when the dataset is distributed stored among different machines. Then the proposed model is further extended to the studies of risk genetic factors for Alzheimer's Disease (AD) among different research institutions, integrating a group feature selection framework to rank the top risk SNPs for AD. For the unsupervised learning problem, I propose a highly efficient solver, termed Stochastic Coordinate Coding (SCC), scaling up the optimization of dictionary learning and sparse coding problems. The common issue for the medical imaging research is that the longitudinal features of patients among different time points are beneficial to study together. To further improve the dictionary learning model, I propose a multi-task dictionary learning method, learning the different task simultaneously and utilizing shared and individual dictionary to encode both consistent and changing imaging features.
prostate cancer is most commonly treated with hormonal therapy. The idea behind
hormonal therapy is to reduce androgen production, which prostate cancer cells
require for growth. Recently, the exploration of the synergistic effects of the drugs
used in hormonal therapy has begun. The aim was to build off of these recent
advancements and further refine the synergistic drug model. The advancements I
implement come by addressing biological shortcomings and improving the model’s
internal mechanistic structure. The drug families being modeled, anti-androgens,
and gonadotropin-releasing hormone analogs, interact with androgen production in a
way that is not completely understood in the scientific community. Thus the models
representing the drugs show progress through their ability to capture their effect
on serum androgen. Prostate-specific antigen is the primary biomarker for prostate
cancer and is generally how population models on the subject are validated. Fitting
the model to clinical data and comparing it to other clinical models through the
ability to fit and forecast prostate-specific antigen and serum androgen is how this
improved model achieves validation. The improved model results further suggest that
the drugs’ dynamics should be considered in adaptive therapy for prostate cancer.
The first topic is an embedded feature selection algorithm termed Expectation-Selection-Maximization (ESM) model that can automatically select features while optimizing the parameters for Gaussian Mixture Model. I introduce a relevancy index (RI) revealing the contribution of the feature in the clustering process to assist feature selection. I demonstrate the efficacy of the ESM by studying two synthetic datasets, four benchmark datasets, and an Alzheimer’s Disease dataset.
The second topic focuses on extending the application of ESM algorithm to handle mixed datatypes. The Gaussian mixture model is generalized to Generalized Model of Mixture (GMoM), which can not only handle continuous features, but also binary and nominal features.
The last topic is about Uncertainty Quantification (UQ) of the feature selection. A new algorithm termed ESOM is proposed, which takes the variance information into consideration while conducting feature selection. Also, a set of outliers are generated in the feature selection process to infer the uncertainty in the input data. Finally, the selected features and detected outlier instances are evaluated by visualization comparison.
The main objective of this work is to investigate and improve measurement policies on the basis of quality control in the transduction/expansion bio-manufacturing processes. More specifically, this study addresses the issue of measuring yield within the transduction/expansion phases of gene therapy. To do so, it was decided to model the process as a Markov Decision Process where the decisions being made are optimally chosen to create an overall optimal measurement policy; for a set of predefined parameters.
The effects of urbanization on ozone levels have been widely investigated over cities primarily located in temperate and/or humid regions. In this study, nested WRF-Chem simulations with a finest grid resolution of 1 km are conducted to investigate ozone concentrations O3 due to urbanization within cities in arid/semi-arid environments. First, a method based on a shape preserving Monotonic Cubic Interpolation (MCI) is developed and used to downscale anthropogenic emissions from the 4 km resolution 2005 National Emissions Inventory (NEI05) to the finest model resolution of 1 km. Using the rapidly expanding Phoenix metropolitan region as the area of focus, we demonstrate the proposed MCI method achieves ozone simulation results with appreciably improved correspondence to observations relative to the default interpolation method of the WRF-Chem system. Next, two additional sets of experiments are conducted, with the recommended MCI approach, to examine impacts of urbanization on ozone production: (1) the urban land cover is included (i.e., urbanization experiments) and, (2) the urban land cover is replaced with the region's native shrubland. Impacts due to the presence of the built environment on O3 are highly heterogeneous across the metropolitan area. Increased near surface O3 due to urbanization of 10–20 ppb is predominantly a nighttime phenomenon while simulated impacts during daytime are negligible. Urbanization narrows the daily O3 range (by virtue of increasing nighttime minima), an impact largely due to the region's urban heat island. Our results demonstrate the importance of the MCI method for accurate representation of the diurnal profile of ozone, and highlight its utility for high-resolution air quality simulations for urban areas.
Forecasts of noise pollution from a highway line segment noise source are obtained from a sound propagation model utilizing effective sound speed profiles derived from a Numerical Weather Prediction (NWP) limited area forecast with 1 km horizontal resolution and near-ground vertical resolution finer than 20 m. Methods for temporal along with horizontal and vertical spatial nesting are demonstrated within the NWP model for maintaining forecast feasibility. It is shown that vertical nesting can improve the prediction of finer structures in near-ground temperature and velocity profiles, such as morning temperature inversions and low level jet-like features. Accurate representation of these features is shown to be important for modeling sound refraction phenomena and for enabling accurate noise assessment. Comparisons are made using the parabolic equation model for predictions with profiles derived from NWP simulations and from field experiment observations during mornings on November 7 and 8, 2006 in Phoenix, Arizona. The challenges faced in simulating accurate meteorological profiles at high resolution for sound propagation applications are highlighted and areas for possible improvement are discussed.
Physical mechanisms of incongruency between observations and Weather Research and Forecasting (WRF) Model predictions are examined. Limitations of evaluation are constrained by (i) parameterizations of model physics, (ii) parameterizations of input data, (iii) model resolution, and (iv) flux observation resolution. Observations from a new 22.1-m flux tower situated within a residential neighborhood in Phoenix, Arizona, are utilized to evaluate the ability of the urbanized WRF to resolve finescale surface energy balance (SEB) when using the urban classes derived from the 30-m-resolution National Land Cover Database. Modeled SEB response to a large seasonal variation of net radiation forcing was tested during synoptically quiescent periods of high pressure in winter 2011 and premonsoon summer 2012. Results are presented from simulations employing five nested domains down to 333-m horizontal resolution. A comparative analysis of model cases testing parameterization of physical processes was done using four configurations of urban parameterization for the bulk urban scheme versus three representations with the Urban Canopy Model (UCM) scheme, and also for two types of planetary boundary layer parameterization: the local Mellor–Yamada–Janjić scheme and the nonlocal Yonsei University scheme. Diurnal variation in SEB constituent fluxes is examined in relation to surface-layer stability and modeled diagnostic variables. Improvement is found when adapting UCM for Phoenix with reduced errors in the SEB components. Finer model resolution is seen to have insignificant (<1 standard deviation) influence on mean absolute percent difference of 30-min diurnal mean SEB terms.