1,721,072 research outputs found
Data analytics and optimization methods in biomedical systems: from microbes to humans
Get PDFData analytics and optimization theory are well-developed techniques to describe, predict and optimize real-world systems, and they have been widely used in engineering and science. This dissertation focuses on applications in biomedical systems, ranging from the scale of microbial communities to problems relating to human disease and health care.
Starting from the microbial level, the first problem considered is to design metabolic division of labor in microbial communities. Given a number of microbial species living in a community, the starting point of the analysis is a list of all metabolic reactions present in the community, expressed in terms of the metabolite proportions involved in each reaction. Leveraging tools from Flux Balance Analysis (FBA), the problem is formulated as a Mixed Integer Program (MIP) and new methods are developed to solve large scale instances. The strategies found reveal a large space of nuanced and non-intuitive metabolic division of labor opportunities, including, for example, splitting the Tricarboxylic Acid Cycle (TCA) cycle into two separate halves. More broadly, the landscape of possible 1-, 2-, and 3-strain solutions is systematically mapped at increasingly tight constraints on the number of allowed reactions.
The second problem addressed involves the prediction and prevention of short-term (30-day) hospital re-admissions. To develop predictive models, a variety of classification algorithms are adapted and coupled with robust (regularized) learning and heuristic feature selection approaches. Using real, large datasets, these methods are shown to reliably predict re-admissions of patients undergoing general surgery, within 30-days of discharge. Beyond predictions, a novel prescriptive method is developed that computes specific control actions with the effect of altering the outcome. This method, termed Prescriptive Support Vector Machines (PSVM), is based on an underlying SVM classifier. Applied to the hospital re-admission data, it is shown to reduce 30-day re-admissions after surgery through better control of the patient’s pre-operative condition. Specifically, using the new method the patient’s pre-operative hematocrit is regulated through limited blood transfusion.
In the last problem in this dissertation, a framework for parameter estimation in Regularized Mixed Linear Regression (MLR) problems is developed. In the specific MLR setting considered, training data are generated from a mixture of distinct linear models (or clusters) and the task is to identify the corresponding coefficient vectors. The problem is formulated as a Mixed Integer Program (MIP) subject to regularization constraints on the coefficient vectors. A number of results on the convergence of parameter estimates for MLR are established. In addition, experimental prediction results are presented comparing the prediction algorithm with mean absolute error regression and random forest regression, in terms of both accuracy and interpretability
Robust learning to rank models and their biomedical applications
Get PDFThere exist many real-world applications such as recommendation systems, document retrieval, and computational biology where the correct ordering of instances is of equal or greater importance than predicting the exact value of some discrete or continuous outcome. Learning-to-Rank (LTR) refers to a group of algorithms that apply machine learning techniques to tackle these ranking problems. Despite their empirical success, most existing LTR models are not built to be robust to errors in labeling or annotation, distributional data shift, or adversarial data perturbations. To fill this gap, we develop four LTR frameworks that are robust to various types of perturbations. First, Pairwise Elastic Net Regression Ranking (PENRR) is an
elastic-net-based regression method for drug sensitivity prediction. PENRR infers robust predictors of drug responses from patient genomic information. The special design of this model (comparing each drug with other drugs in the same cell line and comparing that drug with itself in other cell lines) significantly enhances the accuracy of the drug prediction model under limited data. This approach is also able to solve the problem of fitting on the insensitive drugs that is commonly encountered in regression-based models. Second, Regression-based Ranking by Pairwise Cluster Comparisons (RRPCC) is a ridge-regression-based method for ranking clusters of similar protein complex conformations generated by an underlying docking program (i.e., ClusPro). Rather than using regression to predict scores, which would equally penalize deviations for either low-quality and high-quality clusters, we seek to predict the difference of scores for any pair of clusters corresponding to the same complex. RRPCC combines these pairwise assessments to form a ranked list of clusters, from higher to lower quality. We apply RRPCC to clusters produced by the automated docking server ClusPro and, depending on the training/validation strategy, we show. improvement by 24%–100% in ranking acceptable or better quality clusters first, and by 15%–100% in ranking medium or better quality clusters first. Third, Distributionally Robust Multi-Output Regression Ranking (DRMRR) is a listwise LTR model that induces robustness into LTR problems using the Distributionally Robust Optimization framework. Contrasting to existing methods, the scoring function of DRMRR was designed as a multivariate mapping from a feature vector to a vector of deviation scores, which captures local context information and cross-document interactions. DRMRR employs ranking metrics (i.e., NDCG) in its output. Particularly, we used the notion of position deviation to define a vector of relevance score instead of a scalar one. We then adopted the DRO framework to minimize a worst-case expected multi-output loss function over a probabilistic ambiguity set that is defined by the Wasserstein metric. We also presented an equivalent convex reformulation of the DRO problem, which is shown to be tighter than the ones proposed by the previous studies. Fourth, Inversion Transformer-based Neural Ranking (ITNR) is a Transformer-based model to predict drug responses using RNAseq gene expression profiles, drug descriptors, and drug fingerprints. It utilizes a Context-Aware-Transformer architecture as its scoring function that ensures the modeling of inter-item dependencies. We also introduced a new loss function using the concept of Inversion and approximate permutation matrices. The accuracy and robustness of these LTR models are verified through three medical applications, namely cluster ranking in protein-protein docking, medical document retrieval, and drug response prediction
Machine learning for effective predictions and prescriptions in health care
Get PDFEarly detection of acute hospitalizations and enhancing treatment efficiency is important to improve patients’ long-term life quality and reduce health care costs. This thesis develops data-driven methods to predict important health related events and optimize treatment options. Applications include predicting chronic-disease-related hospitalizations, predicting the effect of interventions, such as In Vitro Fertilization (IVF), and learning and improving upon physicians' prescription policies.
For a binary hospitalization classification problem, and to strike a balance between accuracy and interpretability of the prediction, a novel Alternating Clustering and Classification (ACC) method is proposed, which employs an alternating optimization approach that jointly identifies hidden patient clusters and adapts classifiers to each cluster. Convergence and out-of-sample guarantees for this algorithm are established. The algorithm is validated on large data sets from the Boston Medical Center, the largest safety-net hospital system in New England.
For the IVF outcome prediction problem, and for women who have difficulty conceiving, several predictive models that estimate IVF success rate are designed. For predicted non-pregnant subjects, an algorithm further predicts whether no embryos were implanted (due to embryo abnormalities) or pregnancy did not occur despite implantation. Results are presented to assess the sensitivity of the models to specific predictive variables.
The third problem considered amounts to modeling the patients' disease progression as a Markov Decision Process (MDP), and seeking to estimate the physicians' prescription policy and the disease state transition probabilities. Two regularized maximum likelihood estimation algorithms for learning the transition probability model and policy, respectively, are proposed. A sample complexity result that guarantees a low regret with a relatively small amount of training samples is established. The theoretical results are illustrated using a healthcare example.
Finally, the thesis develops a framework for learning and improving the pharmacological therapy algorithm used by physicians to treat type 2 diabetes, based on prescription data. First, the proposed approach predicts the outcomes of prescriptions using regression, and then a policy consistent with physicians' prescriptions using a parametric multi-class classification method is synthesized from data. Then, by optimizing over algorithm parameters in the prescription model,
the algorithm is able to achieve better glucose control effects
Load flow and optimal power flow in power distribution systems - application of Riemannian optimization and holomorphic embedding
Get PDFDistribution networks are undergoing unprecedented challenges originated from the large-scale adoption of distributed energy resources, price responsive demand, electric storage resources, electric vehicles, etc. Power system analysis techniques such as Load Flow (LF) and Optimal Power Flow (OPF) are necessary to ensure secure and optimal operation in an increasingly active, distributed, and dynamic distribution grid operation. This dissertation presents robust and computationally efficient solution techniques for LF and OPF problems.
In the first part of this dissertation, we utilize Riemannian optimization and present two solution methods. The first solution method is applicable to the LF problem and is shown to fall into the category of Riemannian approximate Newton methods, which guarantees descent at each iteration while maintaining a local superlinear convergence rate. The second solution method is a Riemannian Augmented Lagrangian Method (RALM) which applies to the OPF problem. The proposed solution approach exploits the geometrical properties of the power flow equations and ensures the physical feasibility of the solution. Computational experiments on several distribution networks provide encouraging results in terms of solution quality and speed.
The second part of this dissertation employs Holomorphic embedding methods and presents two LF solution techniques. The first solution technique improves the computational efficiency of the Holomorphic Embedding Load flow Method (HELM). Numerical experimentation demonstrates overall time savings of up to 30% on IEEE radial distribution test cases. The second solution technique extends HELM to three-phase distribution networks with a generic topology and wye/delta connected ZIP load models. We demonstrate the efficacy of the proposed method through numerical results.
The third part of this dissertation is focused on exploiting the existing network infrastructure to improve the system operation. We present an optimal line switching and bus splitting heuristic considering AC and N-1 contingency constraints and apply the proposed method to several IEEE standard test networks. We also provide directions for possible extensions to distribution networks. The proposed method identifies a network topology that reduces the operation cost while maintaining AC feasibility and initial system reliability level in the sense of N-1 contingency requirements
Distributed solution techniques to regulate the load consumption of a residential neighborhood
No full text2025The main challenge utility companies face is to meet the electricity demand during peak hours, and residential dwellings significantly contribute to peak load, with space heating and cooling, water heating, electric vehicle charging, and the use of other routine appliances. Demand Response (DR) programs aim to control these loads to facilitate matching load demand with available supply, and various demand response strategies have been proposed for residential communities. However, their use in real-world scenarios can be limited due to computational challenges and privacy concerns. In order to overcome these limitations, developing distributed solution techniques is as important as designing demand response strategies. This dissertation focuses on DR strategies for residential communities and designs distributed solutions techniques. The continuous participation of users is a key factor to the success of demand response programs, and methods failing to maintain user comfort may not realize their full potential. Hence, the first part of the dissertation proposes two residential demand response strategies based on model predictive control algorithms to ensure user comfort. The first algorithm employs a gradient-based distributed optimization approach, while the second utilizes a Dantzig-Wolfe decomposition technique to solve the problem in a distributed manner. Our research demonstrates the effectiveness of our distributed solution strategies in solving the formulated problem within a simulated environment, compared to a commercial solver attempting to solve the centralized version of the problem. Deep Reinforcement Learning (RL) has shown its potential to solve complex sequential decision-making problems, and it is a prominent tool for DR-related applications due to its model-free nature. However, RL algorithms can suffer from sample inefficiency issues, which limit their use in real-world applications. To address this issue, the second part of the dissertation proposes utilizing expert data to improve the performance of RL algorithms. In addition to proposing a general framework that improves the efficiency of RL algorithms by leveraging expert data, we investigate the effect of utilizing expert data in a demand response related RL algorithm
Transforming dementia diagnosis & prognosis through AI
No full text2025Dementia, characterized by a progressive decline in cognitive function, poses significant societal challenges. This dissertation develops a suite of Artificial Intelligence (AI) approaches that analyze digital data from subjects to assess and predict cognitive decline associated with Alzheimer’s disease and related dementias.First, a deep convolutional neural network (CNN) was applied to images derived from the Clock Drawing Test (CDT), using transfer learning to extract high-level features. Combined with demographic information such as age, the model demonstrated strong performance, yielding high accuracy and sensitivity in identifying cognitive impairment.
Next, language models were employed to automate the diagnosis of cognitive impairment from Electronic Health Records (EHR). Clinical notes were analyzed with the Universal Sentence Encoder, focusing on sentences containing keywords related to dementia and daily activities. The analysis involved multiple binary classification tasks, where the language model generated text features from random and encounter-based sampling methods, achieving high performance in detecting both dementia and mild cognitive impairment (MCI).
A novel natural language processing (NLP) approach was also developed to identify different stages of dementia using automated transcriptions of voice recordings from neuropsychological (NP) tests. The pipeline incorporated speech recognition, speech diarization, transformer-based sentence encoding, and logistic regression models. This method achieved strong predictive performance for distinguishing between different cognitive stages, from normal cognition to dementia and MCI. Building on the voice data, this approach was extended to predict the progression from MCI to Alzheimer’s disease, leveraging speech features alongside demographic factors such as age and education. The results demonstrated promising accuracy and sensitivity, outperforming models using traditional NP test scores.
Finally, smartphone-based cognitive assessments were evaluated for their utility in early- stage detection of cognitive impairment. The approach focuses on text features derived from voice recordings of cognitive tests and embedding them using language models. The proposed method, which combines text features with age data, demonstrates high predictive accuracy, particularly in higher-risk populations.
Through the integration of digital technologies, smartphone-based tests, machine learning, NLP, and computer vision, this research offers accessible and reliable tools for dementia diagnosis and prognosis, with potential for broad clinical applications and early interventions
Distributionally Robust Learning under the Wasserstein Metric
Get PDFThis dissertation develops a comprehensive statistical learning framework that is robust to (distributional) perturbations in the data using Distributionally Robust Optimization (DRO) under the Wasserstein metric. The learning problems that are studied include: (i) Distributionally Robust Linear Regression (DRLR), which estimates a robustified linear regression plane by minimizing the worst-case expected absolute loss over a probabilistic ambiguity set characterized by the Wasserstein metric; (ii) Groupwise Wasserstein Grouped LASSO (GWGL), which aims at inducing sparsity at a group level when there exists a predefined grouping structure for the predictors, through defining a specially structured Wasserstein metric for DRO; (iii) Optimal decision making using DRLR informed K-Nearest Neighbors (K-NN) estimation, which selects among a set of actions the optimal one through predicting the outcome under each action using K-NN with a distance metric weighted by the DRLR solution; and (iv) Distributionally Robust Multivariate Learning, which solves a DRO problem with a multi-dimensional response/label vector, as in Multivariate Linear Regression (MLR) and Multiclass Logistic Regression (MLG), generalizing the univariate response model addressed in DRLR. A tractable DRO relaxation for each problem is being derived, establishing a connection between robustness and regularization, and obtaining upper bounds on the prediction and estimation errors of the solution. The accuracy and robustness of the estimator is verified through a series of synthetic and real data experiments. The experiments with real data are all associated with various health informatics applications, an application area which motivated the work in this dissertation. In addition to estimation (regression and classification), this dissertation also considers outlier detection applications
Data science approaches to data center sustainability and transportation predictive analytics
Get PDFToday's plethora of available data has brought about technological advancement in multiple facets of life. One area that has burgeoned is the development of intelligent cities that leverage technology to deliver improved quality of life. The concept of smart cities involves the incorporation of digital intelligence to urban systems with the hope that improvements in specific functions of a city can, all together, yield benefits in terms of resident satisfaction and the broader societal goals of efficiency and sustainability. In this dissertation, we study the advancement of smart cities through the development of the sub-areas of smart energy and transportation.
Regarding smart energy, we focus on data center sustainability and data center inclusion in the smart grid to extract flexibility and ease congestion in various scenarios. We leverage queuing system theory along with optimization techniques to model data centers and propose demand response models in various settings where we highlight the benefits of data center inclusion in the smart grid. We first propose frameworks for data center interaction with an aggregator that attempts to extract flexibility from the data centers via a price incentive in a network-less setting. We further consider this setting for the case where data centers have the ability to service different types of jobs. Finally, we consider the network setting where data centers are included in the transmission grid with other loads and generators and highlight their ability to provide benefits to the grid.
Regarding smart transportation, we focus on predictive analytics for transportation using machine learning and deep learning approaches. Specifically, we propose state-of-the-art transformer-based architectures for the task of destination prediction using artificial and real human movement data in both indoor and outdoor settings. Destination prediction involves the use of movement data to create intelligent models that can predict intended destinations given partial trajectory information. The use of such a system that knows in advance where a user intends to travel to can offer many functionalities. In an outdoor setting, tailored advertisements, adjacent points of interest, and open parking spots to a destination can be broadcasted to a user. In an indoor setting, a destination prediction system can allow for smart elevator control in large multi-floor buildings and the creation of density-based security systems that focus on high interest areas in a building.
The contributions of this dissertation serve as strong building blocks for the development of smart cities and lead to multiple exciting directions for future work.2025-01-16T00:00:00
Towards effective and robust transformer-based models for biomedical applications
No full text2025Transformer-based deep neural networks have achieved great success in biomedical research and practice. Leveraging the strength of the self-attention mechanism and self-supervised learning schemes such as Masked Language Modeling (MLM), transformer-based Language Models (LMs) like Clinical BERT have shown impressive performance on clinical Natural Language Processing (NLP) tasks, while Vision Transformers (ViTs) also improve the traditional Convolutional Neural Network (CNN) models on clinical image analysis. These successes indicate the general sequence modeling power of transformers, which can be potentially extended to effectively analyze the biomedical sequential data in other modalities. On the other hand, ViTs trained under Empirical Risk Minimization (ERM) are known to be vulnerable to perturbations in the image, which limits their applications in biomedicine. This dissertation focuses on developing effective and robust transformer models and learning methods for a variety of biomedical applications. We propose a pre-training method to introduce prior knowledge from a medical knowledge base to transformer-based LMs, which can be implemented jointly with traditional MLM and improve the LM's ability on clinical text understanding. Inspired by the strength of MLM, we pre-train and fine-tune transformer-based protein LMs to achieve state-of-the-art MHC-peptide binding predictions. We also propose a causal Electronic Health Record (EHR) modeling scheme for general structured EHR modeling, which is implemented to train a Generative Pre-trained Transformer (GPT) and support real-world, unsupervised novel disease detection. To improve the robustness of transformers, we develop a distributionally robust deep learning framework, which significantly reduces the ViT classifier's error rate under image perturbations, and improves the stroke diagnosis accuracy under accelerated MRI settings. Our work not only practically improves the applicability and reliability of Artificial Intelligence (AI) in healthcare, but also provides general effective and robust sequence data processing frameworks for other research domains
Data driven models in health care management
Get PDFThe increasing role of Machine Learning (ML) in scientific and biomedical fields has significantly advanced health and medical informatics. ML-based predictive models have shown great potential in identifying critical features and enabling early-stage treatment. This research explores the increasing worldwide shift towards using data-driven models to support decision-making in healthcare.
This work investigates data-driven techniques to predict health-related events and elucidates important predictive variables. Firstly, it identifies key factors in predicting poorly controlled hypertension using demographic and socioeconomic data, revealing that age, race, social determinants of health, mental health, marital status, cigarette use, and gender are predictive of high systolic blood pressure (SBP). Secondly, it develops a personalized hypertension management model that optimizes medication outcomes using Distributionally Robust Optimization (DRO) regularized regression and K-Nearest Neighbors (K-NN) regression, achieving a significant reduction in SBP compared to standard care. Thirdly, it predicts critical COVID-19 treatment outcomes including hospitalization, ICU care, mechanical ventilation, and mortality using both linear and nonlinear classification methods, achieving high predictive accuracy. Lastly, it emphasizes the importance of model interpretability and alignment with medical knowledge to ensure adoption and trust in clinical settings. By introducing the Wasserstein DRO formulation and the Grouped LASSO (GLASSO) algorithm, the research demonstrates enhanced interpretability and credibility, effectively merging model performance with expert knowledge.
Together, these contributions demonstrate the potential of data-driven models to enhance healthcare delivery and patient care by providing a comprehensive approach that balances interpretability, credibility, and expert insight integration
- …
