Search

Early disease diagnosis can significantly improve patient survival rates as appropriate treatment strategies can be timely administered. A promising approach for disease diagnosis is to analyse chemical biomarkers present in bodily fluids as these molecules can provide insights into human metabolic and physiological processes. Changes in the identity and concentrations of such chemicals can help distinguish healthy from disease states. However, some current methods used to collect, analyse, and identify these chemicals have been challenged by limitations in sampling protocols, the resolving power of instruments, and the ability to interpret advanced data analysis methods. This thesis comprises of five concurrent efforts to enhance diagnostic accuracy by investigating various machine learning and analytical approaches. Firstly, an interpretable machine learning framework for binary disease classification is presented. Using this framework on blood plasma and skin sebum data, the diagnostic performance for Parkinson’s disease and key disease biomarkers are reported. Secondly, a protocol and recommendations for robust skin sebum analysis is described. Following a semi-longitudinal study, the various factors that can impact the collection and detection of volatile organic compounds present in skin sebum is discussed. Thirdly, the clinical utility of high-field asymmetric waveform ion mobility spectrometry (FAIMS) for disease diagnosis is reported. Based on a systematic review and meta-analysis, the diagnostic accuracy and clinical implications of using FAIMS is discussed. Fourthly, the performance of high-resolution FAIMS resulting in enhanced ion separations is reported. Using high-resolution FAIMS, the fundamentals that govern the separation of protonation protein isomers is described. Finally, the use of high-resolution FAIMS to analyse volatile organic compounds present in exhaled breath is demonstrated. Using atmospheric pressure chemical ionisation coupled with high-resolution FAIMS, untargeted breath analysis on individual breath profiles is reported. Overall, by improving analytical and machine learning methods, these findings should increase diagnostic accuracy and enable greater confidence in biomarker analysis.

Antimicrobial stewardship programs integrate multiple disciples that aim to preserve the effectiveness and availability of antimicrobial drugs, through judicious antimicrobial use, proper disease diagnosis, and medical decision-making while preserving animal health and welfare. This dissertation presents different strategies to improve antimicrobial stewardship on dairy farms using educational, epidemiological, and genomic approaches. My doctoral dissertation projects have common focus research on antimicrobial stewardship and antimicrobial resistance on the dairy farm and are divided into four main strategic areas: 1) Development and evaluation of on-farm animal health and diagnostic training program for farm workers; 2) Use of a scoping review to improve accuracy and consensus for diagnosis of metritis in dairy cows; 3) Genomic screening of Escherichia coli form cows with metritis, with a focus on AMR and virulence genes; 4) Reducing the consequential effect of antimicrobial use on dairy farms through novel approaches to degrade antimicrobials in waste. The findings of this dissertation highlight the current disparity in robust and clear criteria used to diagnose puerperal metritis in literature, with a considerable variation in the clinical signs of disease used for diagnosis and consequent treatment decisions; the importance of educational training programs targeting farmworkers to improve antimicrobial stewardship knowledge, that considers individual factors, farm size management practices, among other factors. This dissertation also contributes to the understanding of the role of Escherichia coli in the pathogenesis of metritis in dairy cattle, showing a high genetic diversity in the pangenome. Lastly, it shows that alkaline pH and heat treatments might be alternative cost-effective on-farm strategies that could increase the degradation of antimicrobials in waste milk.

Abstract: Nowadays, with the rapid development of science and technology, human healthcare has become a hot topic and attracts more and more attention. A lot of researchers work on different technologies to contribute to our healthcare no matter disease diagnosis and prognosis or disease treatment. In this thesis, we first introduce two different automated diseases diagnosis approaches and then design a novel gene delivery system that can help treat genetic diseases. Artificial intelligence (AI) is a popular research topic now and a lot of researchers are working on it. AI has various successful applications in computer vision, automatic speech recognition, natural language processing, audio recognition, bioinformatics and has been proven to be used in disease diagnosis. Two automated diseases diagnosis approaches are designed for depression and tuberculosis using different machine learning (ML) algorithms, respectively. Depression is one of the most common mental disorders, and rates of depression in individuals continuously increase each year. Traditional diagnosis methods are mostly based on the professional judgment of mental health, which is prone to individual bias. Therefore, it is crucial to design an effective and robust model for automated depression detection. I proposed a multimodal fusion model comprised of text, audio, and video for both depression detection and assessment tasks. For the text modality, a pre-trained sentence embedding algorithm was utilized to extract semantic representation along with bidirectional Long Short-Term Memory (BiLSTM) to predict depression. We also used principal component analysis (PCA) to reduce the dimensionality of the input feature space and fed it into a support vector machine (SVM) to predict depression based on audio modality. For the video modality, XGBoost was employed to conduct both feature selection and depression detection. The final predictions were given by outputs of different modalities with an ensemble voting algorithm. Experiments on the Distress Analysis Interview Corpus Wizard-of-Oz (DAIC-WOZ) dataset showed our proposed model outperforms the baseline in both depression detection and assessment tasks and has comparable performance with other existing state-of-the-art depression detection methods. Tuberculosis (TB) is a major public health burden affecting about a quarter of the world’s population annually according to the World Health Organization(WHO). Among many control steps, early diagnosis and treatment of TB are critical. Commonly used diagnostic techniques, such as X-ray, TB culture test, TB skin test, and Sputum acid-fast bacillus, have their major limitations. Therefore, new cost-effective diagnostic methods are urgently needed. In this thesis, we used high-resolution liquid chromatography-mass spectrometry (LC-MS) to screen 191 blood samples and discovered kynurenine (Kyn), tryptophan (Trp) and their ratio, Indoleamine 2, 3-dioxygenase (IDO) are excellent TB biomarkers. We employed the logistic regression algorithm to detect pulmonary TB and got excellent performance for classifying health control (HC) vs active tuberculosis (ATB) and latent tuberculosis infection (LTBI) vs ATB. When we used IDO and t-spot to distinguish between nontuberculous lung disease (NTB) and ATB, the results are always satisfying both on the validation set and external independent cohort. For the gene delivery system, we synergistically combine non-viral chemical materials, magnetic nanoparticles (MNPs), and physical technique, low-intensity pulsed ultrasound (LIPUS), to achieve efficiently and targeted gene delivery. The MNPs are iron oxide super-paramagnetic nanoparticles, coated with polyethyleneimine (PEI) giving a highly positively charged surface, which is favorable for the binding of genetic materials. Driven by the paramagnetic properties of the MNPs, the application of an external magnetic field increases transfection efficiency while LIPUS stimulation enhances cell viability and permeability. By combining the effect of the external magnetic field and LIPUS, the genetic material (GFP or Cherry Red plasmid) can enter the cells. The flow cytometry results showed that by using just a magnetic field to direct the genetic material, the transfection efficiency on HEK 293 cells that were treated by our MNPs coupled with LIPUS stimulation, increased a lot and was much higher than the positive control (Lipofectamine 2000) and showed less toxicity. Cell viability after transfection was greatly promoted compared to the standard transfection technique.

Fungi play important roles in various environments. Some of them infect plants and cause economically important diseases. However, many fungal pathogens cause similar symptoms or are even spread asymptomatically, making it difficult to identify them morphologically. Therefore, culture-independent, sequence-based diagnostic methods that can detect and identify fungi independently of the symptoms that they cause are desirable. Whole genome metagenomic sequencing has the potential to enable rapid diagnosis of plant diseases without culturing pathogens and designing pathogen-specific probes. In my study, the MinION nanopore sequencer, a portable single‐molecule sequencing platform developed by Oxford Nanopore Technologies, was employed to detect the fungus Calonectria pseudonaviculata (Cps), the causal agent of the devastating boxwood blight disease of the popular ornamental boxwood (Buxus spp.). Various DNA extraction methods and computational tools were compared. Detection was sensitive with an extremely low false positive rate for most methods. Therefore, metagenomic sequencing is a promising technology that could be implemented in routine diagnostics of fungal diseases. Other fungi may play important roles in the atmosphere because of their ice nucleation activity (INA). INA is the capacity of some particles to induce ice formation above the temperature that pure water freezes (-38°C). Importantly, INPs affect the ratio of ice crystals to liquid droplets in clouds, which in turn affects Earth's radiation balance and the intensity and frequency of precipitation. A few fungal species can produce ice nucleating particles (INPs) that cause ice formation at temperatures ≥ –10°C and they may be present in clouds. Two such fungal genera are Fusarium and Mortierella but little is known about their INPs and the genetic basis of their INA. In my study, F. avenaceum and M. alpina were examined in detail. INPs of both species were characterized and it was found that strains within both species varied in regards to the strength of INA. Whole genome sequencing and comparative genomic studies were then performed to identify putative INA genes. Differential expression analyses at different growth temperatures were also performed. INP properties of the two species shared similarities, both appearing to consist of secreted aggregates larger than 30 kDa. Low temperatures induced INA in both species. Lists of candidate INA genes were identified based on their presence in the strains with the strongest INA and/or induction of their expression at low temperatures and because they either encode secreted proteins or enzymes that produce other molecules known to have INA in other organisms. These genes can now be characterized further to help identify the fungal INA genes in both species. This can be expected to help increase our understanding of the role of fungal INA in the atmosphere.

This dissertation addresses two statistical problems of dealing with noisy data with the aid of additional knowledge. My purpose is to highlight that in the era of big data, there is an increasing number of complicated problems with low signal-to-noise ratio, which cannot be simply solved by existing statistical or machine learning methods. For instance, biological data is notorious for its limited sample size but a substantial number of features (a typical p ≫ n problem). Fortunately, there is always additional knowledge from experts or insights that can be employed to devise smart methods to tackle these noisy data.Chapter 2 discusses my work supervised by Professor Haiyan Huang on the hierarchical multi-label classification. This project is motivated by automatic disease diagnosis, where we aim to predict the patient’s status with limited samples in each disease. The structural information that depicts the relationship between diseases can mitigate the low signal-to-noise-ratio issue. We introduce a new statistic called multidimensional-local-precision-rate (mLPR) for each object in each class. We show that classification decisions made by simply sorting objects across classes, in the descending order of mLPRs, can in theory ensure the class hierarchy and meanwhile leading to the maximization of CATCH, a pre-defined performance metric related to the area under a hit curve. In practical implementation, we need to estimate mLPRs from data. Ranking the objects across classes in the descending order of estimated mLPRs, however, would not ensure the optimization of CATCH and/or the class hierarchy anymore. In response to this, we introduce a new ranking algorithm called HierRank, which optimizes an empirical version of CATCH defined based on the estimated mLPRs. The ranking results from HierRank are ensured to satisfy the hierarchical constraint. The superior performance of our approach over state-of-art methods in literature is demonstrated with a synthetic dataset and two real datasets.Chapter 3 discusses my work supervised by Professor Peter J. Bickel on the binomial mixture model with the U-shape constraint under the regime that the binomial size m can be relatively large compared to the sample size n. This project is motivated by the GeneFishing method (Liu et al., 2019), whose output is a combination of the parameter of interest and the subsampling noise. To tackle the noise in the output, we utilize the observation that the density of the output has a U shape and model the output with the binomial mixture model under a U shape constraint. We first analyze the estimation of the underlying distribution F in the binomial mixture model under various conditions for F. Equipped with these theoretical understandings, we propose a simple method Ucut to identify the cutoffs of the U shape and recover the underlying distribution based on the Grenander estimator. It has been shown that when m = Ω(n), the identified cutoffs converge at the rate O(n^{−1/3}). The L1 distance between the recovered distribution and the true one decreases at the same rate. To demonstrate the performance, we apply our method to varieties of simulation studies, a GTEX dataset used in (Liu et al., 2019) and a single cell dataset from Tabula Muris.

The intense interest in quantifying the levels of nucleic acid-based biomarker because of its importance as a blood-derived biomarker for early diagnosis of cancer. The analysis of nucleic acid-based biomarker presents unique challenges because of its short half-life, ultralow abundance in the presence of an overwhelming background of circulating cell-free DNA and direct analysis in complex blood samples. Here the concept of dispersible electrodes rises to these challenges. Dispersible electrodes are gold-coated magnetic nanoparticles (Au@MNPs) that are modified with biorecognition molecules against analytes of interest. The Au@MNPs are distributed throughout a sample to capture the target analyte and then brought back to the surface of a macroelectrode under magnetic control to allow electrochemical detection of the amount of the analyte. This idea of breaking the electrode up into tiny nanoscale particles that seek out analytes rather than the conventional paradigm of making the analyte find the sensor allows preconcentration and detection of analytes to occur with the same gold-coated magnetic nanoparticle but at different times, solutions, and interfaces. It leads to reduced diffusional pathlengths, shortened analysis times, lowered detection limit and in-built amplification system. This dissertation presents the implementation of dispersible electrodes sensing system for the rapid and ultrasensitive electrochemical detection of cancer biomarkers. Three biosensors have been successfully developed that can diagnose TP53 gene, which is genetically linked to half of all cancers, and two sequence-specific DNA molecules that indicate the occurrence of lung cancer and ovarian cancer, respectively. Chapter 3 describes an electrochemical biosensor for detecting the tumour suppressor TP53 gene, the most frequently altered gene in human tumours. This sensor could detect concentration of TP53 gene from 1 aM to 10 nM with single-base resolution in blood samples. Chapter 4 describes an electrochemical biosensing method for DNA methylation analysis. DNA methylation as the stable epigenetic biomarker, has the potential to detect ovarian cancer up to two years in advance of clinical diagnosis. The 110 nucleotide DNA sequence containing a single-site methylation with the lowest detected concentration of 2 aM can be detected in blood samples. Chapter 5 describes an electrochemical analysis method for quantifying the amount of circulating tumour DNA (ctDNA) in blood. This sensor can selectively detect short- and long-strand DNA targets with ultralow detection limit of 3.3 aM. Meanwhile, this sensor provides a promising method to achieve 20 min response time and minimally invasive cancer early diagnosis in blood samples. The success of developing these three biosensors means that the cancer can be diagnosed when it is still curable and as earlier as before any biological symptoms show up. Moreover, these sensors complete such measurements in about 20 min with ultralow detection limits, presenting great potential for early cancer diagnosis.

Machine learning (ML) in healthcare has enabled the automatic detection of diseases from medical images or sensors with high accuracy, often outperforming domain experts. Unfortunately, there is a large variance between how such diagnostic aids perform in research settings and in the real-world. This is due to challenges unique to healthcare that, if unaddressed, limit the usefulness of ML-based software when deployed in hospitals. For example, in human subjects research and clinical trials, subject recruitment and data acquisition are involved processes for both patients and healthcare providers; there are several regulatory and cybersecurity requirements to satisfy to ensure that patient care is not compromised in the pursuit of big data. Without abundant data to train ML models, it can be difficult to elicit good performance that also generalizes well on unseen data in clinical practice. Moreover, ML tools in hospitals cannot function independently but must integrate with existing workflows. There are ethical considerations with respect to how these tools influence the decision making of clinicians and whether they encourage an over-reliance on predictions.In this dissertation, we discuss these and other concerns in the context of three focus areas: stroke, respiratory disease, and Parkinson’s disease. We present machine and deep learning pipelines for weakness detection in stroke patients from video, respiratory disease classification from audio of coughs, and gait assessment in Parkinson’s disease with body sensors. In our efforts, we were cognizant of the technical and human challenges of healthcare. We developed models that not only performed well but also could be trained and rigorously evaluated in a data-conscious way. Our ML solutions ranged from simple leave-one-out approaches to data augmentation with generative adversarial nets. Lastly, we show how ML can more effectively aid medical diagnosis when paired with human-centered design. We describe a clinical decision support system for acute stroke, focusing on the development of an intuitive user interface that balances neurologist assessments with the symptom predictions of our models. This dissertation details novel, human-centered ML techniques for disease diagnosis in neurology and pulmonology, highlighting several lessons learned to benefit the field of machine learning in healthcare at large.

New technologies are revolutionizing disease diagnostics. Sensitive determination of disease related biomarkers to which single molecule detection is the ultimate goal, will help doctors to make precise decision and effective treatment on time which is of significant importance to patience’s wellbeing. Recently nanopore-based sensors emerge rapidly as one promising technology in biophysical and bio-analytical studies due to single entity sensitivity and label-free manner. However, so far the main research has been focusing on works at single-pore sensors. The other concern is that it is challenging to quantify target analytes in ultra-low concentrations as it takes time for rare copies of molecules to find and pass through a nanopore that is governed by Brownian motion and electrophoresis. In addition to this, it is still an unsolved issue to achieve single molecule resolution when applying devices of multipores to biosensing studies based on electric signals. This dissertation presents the development of implementation of biosensing devices based on solid-state chips with nanopore arrays for single-molecule detection of proteins. Chapter 1 provides a perspective review over the recent developments of ultrasensitive bioanalytical sensors and the challenges this class of biosensor faces when moving to single molecule quantitative sensors. Chapter 2 provides an introduction to the field of nanopore and nanopore-based biosensing applications. Recent advances in bio-analytical applications are reviewed followed with a summary about the scientific questions to be addressed in this dissertation. Chapter 3 documents the information for chemicals, materials, instrumentations, operational conditions and procedures that have been employed in the following chapters. In relating to the above mentioned questions, Chapter 4 describes the work about deploying a novel sensing paradigm called “Nanopore Blockade Sensors” for single-molecule detection of prostate-specific antigen (PSA) demonstrated here as a model protein analyte. Rare copies of PSA molecules have been detected in buffered solutions and blood samples at chips of 3 × 3 nanopore array, showing comparable performance to that from a commercially available enzyme-linked immunosorbent assay kit. It is demonstrated that the developed sensors are capable of directly detecting PSA from human whole blood, indicating the potential to sense PSA in complex matrix without pre-treatment process. Also studies with nanopore based sensors demand the capability to differentiate chemical functionalities at nanopore inside and outside areas and potentially conduct systematic investigation about contributions from these two surface areas. Therefore, Chapter 5 introduces a material-specific modification strategy and is demonstrated first at microsized planar surfaces. After successful demonstration at planar interfaces, Chapter 6 presents the work about fabrication and sensing performance of devices with gold metallized nanopore arrays. This is the first ever investigation of fabrication of metallic nanopore array where the metal coating is spatially deposited onto nanopore interior walls only and this work sheds light on research into sophisticated fabrication and further optical studies with the fabricated solid-state metallic nanopores. Chapter 7 summarizes the work of this dissertation and provides a viewing about the future prospects with solid-state devices of (metallic) nanopore arrays as sensors in bio-analytical and biomedical applications.

The success of machine learning algorithms generally depends on intermediate data representation, called features that disentangle the hidden factors of variation in data. Moreover, machine learning models are required to be generalized, in order to reduce the specificity or bias toward the training dataset. Unsupervised feature learning is useful in taking advantage of large amount of unlabeled data, which is available to capture these variations. However, learned features are required to capture variational patterns in data space. In this dissertation, unsupervised feature learning with sparsity is investigated for sparse and local feature extraction with application to lung segmentation, interpretable deep models, and Alzheimer's disease classification. Nonnegative Matrix Factorization, Autoencoder and 3D Convolutional Autoencoder are used as architectures or models for unsupervised feature learning. They are investigated along with nonnegativity, sparsity and part-based representation constraints for generalized and transferable feature extraction.

Neutral zone classifiers include 'no-decision' as a classification outcome. This paper extends neutral zone classifiers to sequential contexts for analyzing longitudinal data. Applications could include medical diagnosis where a decision variable is repeatedly measured on each subject with the expectation of being able to ultimately identify a patient disease status. Sequential classifiers monitor the sequence of measurements and decide when to stop sampling and how to classify the subject. Bayesian sequential classification rules make classifications on the basis of the minimum expected loss. This approach is a challenge due to computational complexity associated with evaluating the expected future costs. We consider Gaussian contexts. In the homogeneous case we demonstrate the equivalence between sequential Bayesian classifier and a simpler boundary-based framework. A solution for the heterogeneous Gaussian case is presented using the boundary-based framework. A recursive algorithm is developed to efficiently determine decision boundaries that minimize the overall expected cost.Alternative decision boundaries which are competitive with the optimal decision boundaries are studied. Misclassification rates and expected sample size are investigated and the results are compared with non-sequential classifiers.

Alzheimer's disease~(AD) is the most frequent neurodegenerative dementia and a growing health problem. Early and accurate diagnosis and prediction of AD is crucial because treatment may be most efficacious if introduced as early as possible. Neuropsychological testing, which is clinically used, sometimes fails to recognize probable dementia, especially to recognize the disease at an early time point such as the mild cognitive impairment~(MCI), which is the prodromal stage of AD. Recently, there has been a realization that magnetic resonance imaging~(MRI) may help diagnoses of AD and MCI. In this dissertation, we introduce an MRI-analysis based algorithm to help diagnose the illness before irreversible neuronal loss has set in, and to help detect brain changes between MCI patients who may convert and may not convert to AD. Given a set of brain MR images, the algorithm first calculates the distance between each pair of images via a registration process. Then images are projected from a high dimensional Euclidean space to a low dimensional Euclidean subspace based on the calculated distances, with a dimension reduction method. Finally classical supervised classification approaches are employed to assign images to appropriate groups in the low dimensional space. The classification accuracy rates we obtained in our experiments are higher than, or at least comparable to, those reported in recently published papers. Moreover, this algorithm can be extended to explore the pathology distribution of AD. Exploring the distribution of AD pathology is of great importance to reveal AD related regional atrophy at specific stages of the disease and provide insight into longitudinal sequence of disease progression. Calculating distances between different brain structures produces different classification accuracy. Those structures yielding higher classification accuracy are considered as pathological regions. Our experimental results on pathology localization are also compared with the reproduced results using other existing popular algorithms; the observations are consistent.

Ichthyobodo necator is a member of the Order Kinetoplastida and an important fish ectoparasite with a broad host and ecological range. When Ichthyobodo were exposed to the anesthetic tricaine, buffered with sodium bicarbonate, parasites remained attached to the skin of fish. When parasites were exposed to unbuffered tricaine, they detached almost completely from the skin. This finding indicated that tricaine should always be buffered when fish are to be clinically evaluated to prevent a reduction or complete loss in parasite load. Unbuffered tricaine was used to collect large numbers of parasites from the skin and gills of hybrid striped bass (Morone saxatilis male x M. chrysops female) for DNA isolation and analysis. A method for obtaining samples for DNA isolation without use of tricaine was also determined. Four preservation methods (Utermohl's solution, ethanol, formalin, freezing) and two DNA isolation methods (DNA isolation kit, crude lysate) were examined. It was concluded that ethanol-fixation was the best preservation method for DNA isolation and PCR amplification of Ichthyobodo DNA. Using these methods, genomic DNA was isolated from Ichthyobodo trophonts collected from five freshwater and one marine fish. The 18S ribosomal RNA (rRNA) gene from each isolate was PCR amplified, cloned and sequenced. The 18S rRNA gene suggested that rather than being a single species, Ichthyobodo represented a complex of at least four different species, each of which occupied a distinct geographical region. The 18S rRNA gene was also amplified and cloned for related kinetoplastid speceis. Phylogenetic relationships within the Order Kinetoplastida (suborders Bodonina and Trypanosomatina) were determined. The suborder Bodonina appeared to contain at least 3 major lineages, none of which correlated with currently recognized families. Ichthyobodo had the most divergent sequence within the Bodonina, indicating it was the most genetically distinct bodonid currently known. The study of Ichthyobodo has contributed to a new understanding of phylogenetics and systematics for the Order Kinetoplastida, as well as insight into the potential for the spread of this parasite locally, nationally and internationally.