Comparison of machine learning approaches and statistical methods for estimation of airborne chemical exposures

Accurate quantitative estimation of exposure using retrospective data has been one of the most challenging tasks in the exposure assessment field. To improve these estimates, some models have been developed using published exposure databases with their corresponding exposure determinants. These models are designed to be applied to reported exposure determinants obtained from study subjects or exposure levels assigned by an industrial hygienist, so quantitative exposure estimates can be obtained. ^ In an effort to improve the prediction accuracy and generalizability of these models, and taking into account that the limitations encountered in previous studies might be due to limitations in the applicability of traditional statistical methods and concepts, the use of computer science- derived data analysis methods, predominantly machine learning approaches, were proposed and explored in this study. ^ The goal of this study was to develop a set of models using decision trees/ensemble and neural networks methods to predict occupational outcomes based on literature-derived databases, and compare, using cross-validation and data splitting techniques, the resulting prediction capacity to that of traditional regression models. Two cases were addressed: the categorical case, where the exposure level was measured as an exposure rating following the American Industrial Hygiene Association guidelines and the continuous case, where the result of the exposure is expressed as a concentration value. Previously developed literature-based exposure databases for 1,1,1 trichloroethane, methylene dichloride and, trichloroethylene were used. ^ When compared to regression estimations, results showed better accuracy of decision trees/ensemble techniques for the categorical case while neural networks were better for estimation of continuous exposure values. Overrepresentation of classes and overfitting were the main causes for poor neural network performance and accuracy. Estimations based on literature-based databases using machine learning techniques might provide an advantage when they are applied to other methodologies that combine `expert inputs' with current exposure measurements, like the Bayesian Decision Analysis tool. The use of machine learning techniques to more accurately estimate exposures from literature-based exposure databases might represent the starting point for the independence from the expert judgment.^

Mouse models and quantitative trait loci of susceptibility to bleomycin- and radiation-induced pulmonary fibrosis

Radiotherapy involving the thoracic cavity and chemotherapy with the drug bleomycin are both dose limited by the development of pulmonary fibrosis. From evidence that there is variation in the population in susceptibility to pulmonary fibrosis, and animal data, it was hypothesized that individual variation in susceptibility to bleomycin-induced, or radiation-induced, pulmonary fibrosis is, in part, genetically controlled. In this thesis a three generation mouse genetic model of C57BL/6J (fibrosis prone) and C3Hf/Kam (fibrosis resistant) mouse strains and F1 and F2 (F1 intercross) progeny derived from the parental strains was developed to investigate the genetic basis of susceptibility to fibrosis. In the bleomycin studies the mice received 100 mg/kg (125 for females) of bleomycin, via mini osmotic pump. The animals were sacrificed at eight weeks following treatment or when their breathing rate indicated respiratory distress. In the radiation studies the mice were given a single dose of 14 or 16 Gy (Co$\sp{60})$ to the whole thorax and were sacrificed when moribund. The phenotype was defined as the percent of fibrosis area in the left lung as quantified with image analysis of histological sections. Quantitative trait loci (QTL) mapping was used to identify the chromosomal location of genes which contribute to susceptibility to bleomycin-induced pulmonary fibrosis in C57BL/6J mice compared to C3Hf/Kam mice and to determine if the QTL's which influence susceptibility to bleomycin-induced lung fibrosis in these progenitor strains could be implicated in susceptibility to radiation-induced lung fibrosis. For bleomycin, a genome wide scan revealed QTL's on chromosome 17, at the MHC, (LOD = 11.7 for males and 7.2 for females) accounting for approximately 21% of the phenotypic variance, and on chromosome 11 (LOD = 4.9), in male mice only, adding 8% of phenotypic variance. The bleomycin QTL on chromosome 17 was also implicated for susceptibility to radiation-induced fibrosis (LOD = 5.0) and contributes 7% of the phenotypic variance in the radiation study. In conclusion, susceptibility to both bleomycin-induced and radiation-induced pulmonary fibrosis are heritable traits, and are influenced by a genetic factor which maps to a genomic region containing the MHC. ^

Mathematical and empirical examinations of some epidemiological procedures

In epidemiology literature, it is often required to investigate the relationships between means where the levels of experiment are actually monotone sets forming a partition on the range of sampling values. With this need, the analysis of these group means is generally performed using classical analysis of variance (ANOVA). However, this method has never been challenged. In this dissertation, we will formulate and present our examination of its validity. First, the classical assumptions of normality and constant variance are not always true. Second, under the null hypothesis of equal means, the test statistic for the classical ANOVA technique is still valid. Third, when the hypothesis of equal means is rejected, the classical analysis techniques for hypotheses of contrasts are not valid. Fourth, under the alternative hypothesis, we can show that the monotone property of levels leads to the conclusion that the means are monotone. Fifth, we propose an appropriate method for handing the data in this situation. ^

Macromolecular interaction studies on FF1-3 of human CA150 and STAT1-importin alpha5 using biophysical and biochemical methods

Macromolecular interactions, such as protein-protein interactions and protein-DNA interactions, play important roles in executing biological functions in cells. However the complexity of such interactions often makes it very challenging to elucidate the structural details of these subjects. In this thesis, two different research strategies were applied on two different two macromolecular systems: X-ray crystallography on three tandem FF domains of transcription regulator CA150 and electron microscopy on STAT1-importin α5 complex. The results from these studies provide novel insights into the function-structure relationships of transcription coupled RNA splicing mediated by CA150 and the nuclear import process of the JAK-STAT signaling pathway. ^ The first project aimed at the protein-protein interaction module FF domain, which often occurs as tandem repeats. Crystallographic structure of the first three FF domains of human CA150 was determined to 2.7 Å resolution. This is the only crystal structure of an FF domain and the only structure on tandem FF domains to date. It revealed a striking connectivity between an FF domain and the next. Peptide binding assay with the potential binding ligand of FF domains was performed using fluorescence polarization. Furthermore, for the first time, FF domains were found to potentially interact with DNA. DNA binding assays were also performed and the results were supportive to this newly proposed functionality of an FF domain. ^ The second project aimed at understanding the molecular mechanism of the nuclear import process of transcription factor STAT1. The first structural model of pSTAT1-importin α5 complex in solution was built from the images of negative staining electron microscopy. Two STAT1 molecules were observed to interact with one molecule of importin α5 in an asymmetric manner. This seems to imply that STAT1 interacts with importin α5 with a novel mechanism that is different from canonical importin α-cargo interactions. Further in vitro binding assays were performed to obtain more details on the pSTAT1-importin α5 interaction. ^

Performance tiers: Implementing comparative effectiveness analysis in the health care setting through operations research and spatial methods

This dissertation develops and tests a comparative effectiveness methodology utilizing a novel approach to the application of Data Envelopment Analysis (DEA) in health studies. The concept of performance tiers (PerT) is introduced as terminology to express a relative risk class for individuals within a peer group and the PerT calculation is implemented with operations research (DEA) and spatial algorithms. The analysis results in the discrimination of the individual data observations into a relative risk classification by the DEA-PerT methodology. The performance of two distance measures, kNN (k-nearest neighbor) and Mahalanobis, was subsequently tested to classify new entrants into the appropriate tier. The methods were applied to subject data for the 14 year old cohort in the Project HeartBeat! study.^ The concepts presented herein represent a paradigm shift in the potential for public health applications to identify and respond to individual health status. The resultant classification scheme provides descriptive, and potentially prescriptive, guidance to assess and implement treatments and strategies to improve the delivery and performance of health systems. ^

A qualitative and quantitative assessment of cancer patients' perceptions of barriers to treatment

This study examined barriers that cancer patients experience in obtaining treatment. The principal aim of the study was to conduct a comprehensive quantitative and qualitative assessment of barriers to cancer treatment for Texas cancer patients. The three specific aims of the study were to: (1) conduct a review and critique of published and unpublished research on barriers to cancer treatment; (2) conduct focus groups for the qualitative assessment of cancer patients' perceived barriers to cancer treatment; and (3) survey a representative sample of cancer patients regarding perceived barriers to treatment. The study was guided by the Aday and Andersen access framework of predisposing, enabling, and need determinants of care-seeking.^ To address the first specific aim, a total of 732 abstracts were examined, from which 154 articles were selected for review. Of these 154 articles, 57 that related directly to research on barriers to cancer treatment were chosen for subsequent analysis. Criteria were applied to each article to evaluate the strength of the study design, sampling and measurement procedures. The major barriers that were consistently documented to influence whether or not cancer patients sought or continued required treatment included problems with communication between the patient and provider, lack of information on side effects, the cost of treatment and associated difficulties in obtaining and maintaining insurance coverage, and the absence of formal and informal networks of social support. Access barriers were generally greater for older, minority women, and patients of lower socioeconomic status.^ To address the second specific aim, a total of eight focus groups (n = 44) were conducted across the State of Texas with cancer patients identified by the Texas Community Oncology Network, American Cancer Society, and community health centers. One important finding was that cost is the greatest hurdle that patients face. Another finding was that with the health care/insurance crisis, an increasing number of physicians are working with their patients to develop individually-tailored payment plans. For people in rural areas, travel to treatment sites is a major barrier due to the travel costs as well as work time forfeited by patients and their family members. A third major finding was the patients' family and church play important roles in providing social and emotional support for cancer patients.^ To address the third aim, a total of 910 cancer patients were surveyed during October and November, 1993. Approximately 65% of the cancer patients responded to the survey. The findings showed that the major barriers to treatment included costs of medications and diagnostic tests, transportation, lack of social support, problems understanding the written information regarding their disease as well as losing coverage or having higher premiums or copayments once they were diagnosed (particularly among blacks).^ Significant differences in reported barriers were found between racial groups. The minority respondents (i.e., blacks and Hispanics) tended to experience more barriers to treatment compared to the white respondents. More specifically, Hispanics were more likely to report transportation as a barrier to treatment than both white and blacks. Future research is needed to better understand the problems that minority cancer patients experience in receiving treatment. (Abstract shortened by UMI.) ^

SOCIO-ENVIRONMENTAL FACTORS ASSOCIATED WITH THE DISTRIBUTION OF DRACUNCULUS MEDINENSIS (GUINEA WORM DISEASE) AND SUGGESTED METHODS FOR CONTROL, IBARAPA, NIGERIA, 1982 (AFRICA)

A population-based cross-sectional survey of socio-environmental factors associated with the prevalence of Dracunculus medinensis (guinea worm disease) was conducted in Idere, a rural agricultural community in Ibarapa, Oyo state, Nigeria, during 1982.^ The epidemiologic data were collected by household interview of all 501 households. The environmental data were collected by analysis of water samples collected from all domestic water sources and rainfall records.^ The specific objectives of this research were to: (a) Describe the prevalence of guinea worm disease in Idere during 1982 by age, sex, area of residence, drinking water source, religion and weekly amount of money spent by the household to collect potable drinking water. (b) Compare the characteristics of cases with non-cases of guinea worm in order to identify factors associated with high risk of infection. (c) Investigate domestic water sources for the distribution of Cyclops. (d) Determine the extent of potable water shortage with a view to identifying factors responsible for such shortage in the community. (e) Describe the effects of guinea worm on school attendance during 1980/1982 school years by class and location of school from piped water supply.^ The findings of this research indicate that during 1982, 31.8 percent of Idere's 6,527 residents experienced guinea worm infection, with higher prevalence of infection recorded in males in their most productive years and females in their teenage years. The role of sex and age to risk of higher infection rate was explained in the context of water related exposure and water intake due to dehydration from physical occupational actitives of subgroups.^ Potable water available to residents was considerably below the minimum recommended by WHO for tropical climates, with sixty-eight percent of water needs of the residents coming from unprotected surface water which harbour Cyclops, the obligatory intermediate host of Dracunculus medinensis. An association was found between periods of relative high density of Cyclops in domestic water and rainfall.^ Impact of guinea worm infection on educational activities was considerable and its implications were discussed, including the implications of the research findings in relation to control of guinea worm disease in Ibarapa. ^

Improving follow-up of abnormal cancer screens using electronic health records: trust but verify test result communication.

BACKGROUND: Early detection of colorectal cancer through timely follow-up of positive Fecal Occult Blood Tests (FOBTs) remains a challenge. In our previous work, we found 40% of positive FOBT results eligible for colonoscopy had no documented response by a treating clinician at two weeks despite procedures for electronic result notification. We determined if technical and/or workflow-related aspects of automated communication in the electronic health record could lead to the lack of response. METHODS: Using both qualitative and quantitative methods, we evaluated positive FOBT communication in the electronic health record of a large, urban facility between May 2008 and March 2009. We identified the source of test result communication breakdown, and developed an intervention to fix the problem. Explicit medical record reviews measured timely follow-up (defined as response within 30 days of positive FOBT) pre- and post-intervention. RESULTS: Data from 11 interviews and tracking information from 490 FOBT alerts revealed that the software intended to alert primary care practitioners (PCPs) of positive FOBT results was not configured correctly and over a third of positive FOBTs were not transmitted to PCPs. Upon correction of the technical problem, lack of timely follow-up decreased immediately from 29.9% to 5.4% (p<0.01) and was sustained at month 4 following the intervention. CONCLUSION: Electronic communication of positive FOBT results should be monitored to avoid limiting colorectal cancer screening benefits. Robust quality assurance and oversight systems are needed to achieve this. Our methods may be useful for others seeking to improve follow-up of FOBTs in their systems.

The role of substance abuse treatment in preventing human immunodeficiency virus transmission among injecting drug users

This exploratory study assesses the utility of substance abuse treatment as a strategy for preventing human immunodeficiency virus (HIV) transmission among injecting drug users (IDUs). Data analyzed in this study were collected in San Antonio, TX, 1989 through 1995 using both qualitative and quantitative methods. Qualitative data included ethnographic interviews with 234 active IDUs; quantitative data included baseline risk assessments and HIV screening plus interviews follow-up interviews administered approximately six months later to 823 IDUs participating in a Federally-funded AIDS community outreach demonstration project.^ Findings that have particularly important implications for substance abuse treatment as an HIV prevention strategy for IDUs are listed below. (1) IDUs who wanted treatment were significantly more likely to be daily heroin users. (2) IDUs who want treatment were significantly more likely to have been to treatment previously. (3) IDUs who wanted treatment at baseline reported significantly higher levels of HIV risk than IDUs who did not want treatment. (4) IDUs who went to treatment between their baseline and follow-up interviews reported significantly higher levels of HIV risk at baseline than IDUs who did not go to treatment. (5) IDUs who went to treatment between their baseline and follow-up interviews reported significantly greater decreases in injection-related HIV risk behaviors. (6) IDUs who went to treatment reported significantly greater decreases in sexual HIV risk behaviors than IDUs who did not go to treatment.^ This study also noted a number of factors that may limit the effectiveness of substance abuse treatment in reducing HIV risk among IDUs. Findings suggest that the impact of methadone maintenance on HIV risk behaviors among opioid dependent IDUs may be limited by the negative manner in which it is perceived by IDUs as well as other elements of society. One consequence of the negative perception of methadone maintenance held by many elements of society may be an unwillingness to provide public funding for an adequate number of methadone maintenance slots. Thus many IDUs who would be willing to enter methadone maintenance are unable to enter it and many IDUs who do enter it are forced to drop out prematurely. ^

DETERMINING THE GENOTYPE-PHENOTYPE CONNECTION IN SYNTHETIC INDUCIBLE GENE EXPRESSION SYSTEMS

Introduction Gene expression is an important process whereby the genotype controls an individual cell’s phenotype. However, even genetically identical cells display a variety of phenotypes, which may be attributed to differences in their environment. Yet, even after controlling for these two factors, individual phenotypes still diverge due to noisy gene expression. Synthetic gene expression systems allow investigators to isolate, control, and measure the effects of noise on cell phenotypes. I used mathematical and computational methods to design, study, and predict the behavior of synthetic gene expression systems in S. cerevisiae, which were affected by noise. Methods I created probabilistic biochemical reaction models from known behaviors of the tetR and rtTA genes, gene products, and their gene architectures. I then simplified these models to account for essential behaviors of gene expression systems. Finally, I used these models to predict behaviors of modified gene expression systems, which were experimentally verified. Results Cell growth, which is often ignored when formulating chemical kinetics models, was essential for understanding gene expression behavior. Models incorporating growth effects were used to explain unexpected reductions in gene expression noise, design a set of gene expression systems with “linear” dose-responses, and quantify the speed with which cells explored their fitness landscapes due to noisy gene expression. Conclusions Models incorporating noisy gene expression and cell division were necessary to design, understand, and predict the behaviors of synthetic gene expression systems. The methods and models developed here will allow investigators to more efficiently design new gene expression systems, and infer gene expression properties of TetR based systems.

Predictive oncology: a review of multidisciplinary, multiscale in silico modeling linking phenotype, morphology and growth.

Empirical evidence and theoretical studies suggest that the phenotype, i.e., cellular- and molecular-scale dynamics, including proliferation rate and adhesiveness due to microenvironmental factors and gene expression that govern tumor growth and invasiveness, also determine gross tumor-scale morphology. It has been difficult to quantify the relative effect of these links on disease progression and prognosis using conventional clinical and experimental methods and observables. As a result, successful individualized treatment of highly malignant and invasive cancers, such as glioblastoma, via surgical resection and chemotherapy cannot be offered and outcomes are generally poor. What is needed is a deterministic, quantifiable method to enable understanding of the connections between phenotype and tumor morphology. Here, we critically assess advantages and disadvantages of recent computational modeling efforts (e.g., continuum, discrete, and cellular automata models) that have pursued this understanding. Based on this assessment, we review a multiscale, i.e., from the molecular to the gross tumor scale, mathematical and computational "first-principle" approach based on mass conservation and other physical laws, such as employed in reaction-diffusion systems. Model variables describe known characteristics of tumor behavior, and parameters and functional relationships across scales are informed from in vitro, in vivo and ex vivo biology. We review the feasibility of this methodology that, once coupled to tumor imaging and tumor biopsy or cell culture data, should enable prediction of tumor growth and therapy outcome through quantification of the relation between the underlying dynamics and morphological characteristics. In particular, morphologic stability analysis of this mathematical model reveals that tumor cell patterning at the tumor-host interface is regulated by cell proliferation, adhesion and other phenotypic characteristics: histopathology information of tumor boundary can be inputted to the mathematical model and used as a phenotype-diagnostic tool to predict collective and individual tumor cell invasion of surrounding tissue. This approach further provides a means to deterministically test effects of novel and hypothetical therapy strategies on tumor behavior.

THE NATURAL AND ORTHOGONAL INTERACTION (NOIA) MODELS FOR QUANTITATIVE TRAITS (QTs) AND COMPLEX DISEASES

My dissertation focuses on developing methods for gene-gene/environment interactions and imprinting effect detections for human complex diseases and quantitative traits. It includes three sections: (1) generalizing the Natural and Orthogonal interaction (NOIA) model for the coding technique originally developed for gene-gene (GxG) interaction and also to reduced models; (2) developing a novel statistical approach that allows for modeling gene-environment (GxE) interactions influencing disease risk, and (3) developing a statistical approach for modeling genetic variants displaying parent-of-origin effects (POEs), such as imprinting. In the past decade, genetic researchers have identified a large number of causal variants for human genetic diseases and traits by single-locus analysis, and interaction has now become a hot topic in the effort to search for the complex network between multiple genes or environmental exposures contributing to the outcome. Epistasis, also known as gene-gene interaction is the departure from additive genetic effects from several genes to a trait, which means that the same alleles of one gene could display different genetic effects under different genetic backgrounds. In this study, we propose to implement the NOIA model for association studies along with interaction for human complex traits and diseases. We compare the performance of the new statistical models we developed and the usual functional model by both simulation study and real data analysis. Both simulation and real data analysis revealed higher power of the NOIA GxG interaction model for detecting both main genetic effects and interaction effects. Through application on a melanoma dataset, we confirmed the previously identified significant regions for melanoma risk at 15q13.1, 16q24.3 and 9p21.3. We also identified potential interactions with these significant regions that contribute to melanoma risk. Based on the NOIA model, we developed a novel statistical approach that allows us to model effects from a genetic factor and binary environmental exposure that are jointly influencing disease risk. Both simulation and real data analyses revealed higher power of the NOIA model for detecting both main genetic effects and interaction effects for both quantitative and binary traits. We also found that estimates of the parameters from logistic regression for binary traits are no longer statistically uncorrelated under the alternative model when there is an association. Applying our novel approach to a lung cancer dataset, we confirmed four SNPs in 5p15 and 15q25 region to be significantly associated with lung cancer risk in Caucasians population: rs2736100, rs402710, rs16969968 and rs8034191. We also validated that rs16969968 and rs8034191 in 15q25 region are significantly interacting with smoking in Caucasian population. Our approach identified the potential interactions of SNP rs2256543 in 6p21 with smoking on contributing to lung cancer risk. Genetic imprinting is the most well-known cause for parent-of-origin effect (POE) whereby a gene is differentially expressed depending on the parental origin of the same alleles. Genetic imprinting affects several human disorders, including diabetes, breast cancer, alcoholism, and obesity. This phenomenon has been shown to be important for normal embryonic development in mammals. Traditional association approaches ignore this important genetic phenomenon. In this study, we propose a NOIA framework for a single locus association study that estimates both main allelic effects and POEs. We develop statistical (Stat-POE) and functional (Func-POE) models, and demonstrate conditions for orthogonality of the Stat-POE model. We conducted simulations for both quantitative and qualitative traits to evaluate the performance of the statistical and functional models with different levels of POEs. Our results showed that the newly proposed Stat-POE model, which ensures orthogonality of variance components if Hardy-Weinberg Equilibrium (HWE) or equal minor and major allele frequencies is satisfied, had greater power for detecting the main allelic additive effect than a Func-POE model, which codes according to allelic substitutions, for both quantitative and qualitative traits. The power for detecting the POE was the same for the Stat-POE and Func-POE models under HWE for quantitative traits.