ADT2R: Adaptive Decision Transformer for Dynamic Treatment Regimes in Sepsis.

Dynamic treatment regimes (DTRs), which comprise a series of decisions taken to select adequate treatments, have attracted considerable attention in the clinical domain, especially from sepsis researchers. Existing sepsis DTR learning studies are mainly based on offline reinforcement learning (RL) approaches working on electronic healthcare records data. However, a trained policy may choose a treatment different from a human clinician's prescription. Furthermore, most of them do not consider: 1) heterogeneity in sepsis; 2) short-term transitions; and 3) the relationship between a patient's health state and the prescription. We propose a novel framework, an adaptive decision transformer for DTR (ADT 2 R), which recommends an optimal treatment action for each time step depending on the heterogeneity of the sepsis and a patient's evolving health states. Specifically, we devise a trajectory-optimization-based module to be trained with supervision for treatments and adaptively aggregate the multihead self-attentions by deliberating on various inherent time-varying patterns among sepsis patients. Furthermore, we estimate the patient's health state by adopting an actor-critic (AC) algorithm and inform the treatment recommendation by learning about its short-term changes. We validated the effectiveness of the proposed framework on the Medical Information Mart for Intensive Care III (MIMIC-III) dataset, an extensive intensive care database, by demonstrating performance comparable to the state-of-the-art methods.

The spatiotemporal dynamics and structural covariates of homicide in the Republic of Korea, 2008-2017: A dynamic spatial panel data approach.

This study examined the relationship between the homicide rate and diverse indicators of social disorganization in the Republic of Korea (South Korea) using datasets collected between 2008 and 2017. Due to the statistical limitations of previous homicide research, which used either cross-sectional or longitudinal methodology, this study applied the dynamic spatial panel data model to explore both the spatial and temporal aspects of homicide. The results demonstrate that the homicide rate is spatially and temporally dependent on those of neighbouring units and the time-lagged homicide rate. Moreover, this study found that divorce rate, unemployment rate, number of males in the neighbourhood and ethnic heterogeneity have a statistically significant impact on the homicide phenomenon. This study contributes to the existing literature by taking a new approach - the dynamic spatial panel data model - to investigate the homicide phenomenon in Korea. In doing so, several suggestions are made for policymakers to respond to homicide rates. Based on the social disorganization theory, these indicators have been found to impact the social network and community members' willingness to engage in social control. This study suggests that customized policies should be implemented to alleviate the level of social disorganization and promote social control over lethal violence.

Multi-View Integrative Attention-Based Deep Representation Learning for Irregular Clinical Time-Series Data.

Electronic health record (EHR) data are sparse and irregular as they are recorded at irregular time intervals, and different clinical variables are measured at each observation point. In this work, to handle irregular multivariate time-series data, we consider the human knowledge of the aspects to be measured and time to measure them in different situations, known as multi-view features, which are indirectly represented in the data. We propose a scheme to realize multi-view features integration learning via a self-attention mechanism. Specifically, we devise a novel multi-integration attention module (MIAM) to extract complex information that is inherent in irregular time-series data. We explicitly learn the relationships among the observed values, missing indicators, and time interval between the consecutive observations in a simultaneous manner. In addition, we build an attention-based decoder as a missing value imputer that helps empower the representation learning of the interrelations among multi-view observations for the prediction task this decoder operates only in the training phase so that the final model is implemented in an imputation-free manner. We validated the effectiveness of our method over the public MIMIC-III and PhysioNet challenge 2012 datasets by comparing with and outperforming the state-of-the-art methods in three downstream tasks i.e., prediction of the in-hospital mortality, prediction of the length of stay, and phenotyping. Moreover, we conduct a layer-wise relevance propagation (LRP) analysis based on case studies to highlight the explainability of the trained model.

Uncertainty-Gated Stochastic Sequential Model for EHR Mortality Prediction.

Electronic health records (EHRs) are characterized as nonstationary, heterogeneous, noisy, and sparse data; therefore, it is challenging to learn the regularities or patterns inherent within them. In particular, sparseness caused mostly by many missing values has attracted the attention of researchers who have attempted to find a better use of all available samples for determining the solution of a primary target task through defining a secondary imputation problem. Methodologically, existing methods, either deterministic or stochastic, have applied different assumptions to impute missing values. However, once the missing values are imputed, most existing methods do not consider the fidelity or confidence of the imputed values in the modeling of downstream tasks. Undoubtedly, an erroneous or improper imputation of missing variables can cause difficulties in the modeling as well as a degraded performance. In this study, we present a novel variational recurrent network that: 1) estimates the distribution of missing variables (e.g., the mean and variance) allowing to represent uncertainty in the imputed values; 2) updates hidden states by explicitly applying fidelity based on a variance of the imputed values during a recurrence (i.e., uncertainty propagation over time); and 3) predicts the possibility of in-hospital mortality. It is noteworthy that our model can conduct these procedures in a single stream and learn all network parameters jointly in an end-to-end manner. We validated the effectiveness of our method using the public data sets of MIMIC-III and PhysioNet challenge 2012 by comparing with and outperforming other state-of-the-art methods for mortality prediction considered in our experiments. In addition, we identified the behavior of the model that well represented the uncertainties for the imputed estimates, which showed a high correlation between the uncertainties and mean absolute error (MAE) scores for imputation.

Modeling regional dynamics in low-frequency fluctuation and its application to Autism spectrum disorder diagnosis.

With the advent of neuroimaging techniques, many studies in the literature have validated the use of resting-state fMRI (rs-fMRI) for understanding functional mechanisms of the brain, as well as for identifying brain disorders or diseases. One of the main streams in recent studies of modeling and analyzing rs-fMRI data is to account for the dynamic characteristics of a brain. In this study, we propose a novel method that directly models the regional temporal BOLD fluctuations in a stochastic manner and estimates the dynamic characteristics in the form of likelihoods. Specifically, we modeled temporal BOLD fluctuation of individual Regions Of Interest (ROIs) by means of Hidden Markov Models (HMMs), and then estimated the 'goodness-of-fit' of each ROI's BOLD signals to the corresponding trained HMM in terms of a likelihood. Using estimated likelihoods of the ROIs over the whole brain as features, we built a classifier that can discriminate subjects with Autism Spectrum Disorder (ASD) from Typically Developing (TD) controls at an individual level. In order to interpret the trained HMMs and a classifier from a neuroscience perspective, we also conducted model analysis. First, we investigated the learned weight coefficients of a classifier by transforming them into activation patterns, from which we could identify the ROIs that are highly associated with ASD and TD groups. Second, we explored the characteristics of temporal BOLD signals in terms of functional networks by clustering them based on sequences of the hidden states decoded with the trained HMMs. We validated the effectiveness of the proposed method by achieving the state-of-the-art performance on the ABIDE dataset and observed insightful patterns related to ASD.

Evaluation of general toxicity and genotoxicity of the silkworm extract powder.

The silkworm extract powder contain 1-deoxynojirimycin (DNJ), a potent α-glycosidase inhibitor, has therapeutic potency against diabetes mellitus. Therefore, natural products containing DNJ from mulberry leaves and silkworm are consumed as health functional food. The present study was performed to evaluate the safety of the silkworm extract powder, a health food which containing the DNJ. The repeated toxicity studies and gentic toxicity studies of the silkworm extract powder were performed to obtain the data for new functional food approval in MFDS. The safety was evaluated by a single-dose oral toxicity study and a 90 day repeated-dose oral toxicity study in Sprague-Dawley rats. The silkworm extract powder was also evaluated for its mutagenic potential in a battery of genetic toxicity test: in vitro bacterial reverse mutation assay, in vitro chromosomal aberration test, and in vivo mouse bone marrow micronucleus assay. The results of the genetic toxicology assays were negative in all of the assays. The approximate lethal dose in single oral dose toxicity study was considered to be higher than 5000 mg/kg in rats. In the 90 day study, the dose levels were wet at 0, 500, 1000, 2000 mg/kg/day, and 10 animals/sex/dose were treated with oral gavage. The parameters that were monitored were clinical signs, body weights, food and water consumptions, ophthalmic examination, urinalysis, hematology, serum biochemistry, necropsy findings, organ weights, and histopathological examination. No adverse effects were observed after the 90 day administration of the silkworm extract powder. The No-Observed-Adverse-Effect-Level (NOAEL) of silkworm extract powder in the 90 day study was 2000 mg/kg/day in both sexes, and no target organ was identified.

Design of a context-aware model to enhance medication adherence.

Medication adherence is important to patients who suffer from chronic disease. Regular medication activity reduces the cost of caring disease and prohibits the worsening of disease condition. To support patients taking medicine correctly, we developed a medication assistance system which alarms medication situation through multimedia messages and help patients to take a medicine. To enable the system copes with various situations related to a medication service, we designed a medication context model and implemented a state based context aware application. We also applied our system to patients and saw a little improvement in medication adherence.

Changes in lipid distribution during aging and its modulation by calorie restriction.

Adipogenesis and ectopic lipid accumulation during aging have a great impact on the aging process and the pathogenesis of chronic diseases with age. However, at present, information on the age-related molecular changes in lipid redistribution patterns and their potential nutritional interventions is sparse. We investigated the mechanism underlying age-related lipid redistribution and its modulation using 5-, 17-, and 24-month-old male Fischer 344 rats fed ad libitum (AL) or a 3-week-long CR (40% less than AL) diet. Results revealed that the activities of adipogenic transcription factors were decreased in the white adipose tissue (WAT) of aged AL rats. In contrast, the skeletal muscle of aged AL rats showed increased fat accumulation through decreased carnitine palmitoyltransferase-1 activity, which was blunted by short-term CR. This study suggests an age-related shift in lipid distribution by reducing the adipogenesis of WAT while increasing intramyocellular lipid accumulation, and that CR can modulate age-related adipogenesis and ectopic lipid accumulation.

Distance correction system for localization based on linear regression and smoothing in ambient intelligence display.

This paper suggests the method of correcting distance between an ambient intelligence display and a user based on linear regression and smoothing method, by which distance information of a user who approaches to the display can he accurately output even in an unanticipated condition using a passive infrared VIR) sensor and an ultrasonic device. The developed system consists of an ambient intelligence display and an ultrasonic transmitter, and a sensor gateway. Each module communicates with each other through RF (Radio frequency) communication. The ambient intelligence display includes an ultrasonic receiver and a PIR sensor for motion detection. In particular, this system selects and processes algorithms such as smoothing or linear regression for current input data processing dynamically through judgment process that is determined using the previous reliable data stored in a queue. In addition, we implemented GUI software with JAVA for real time location tracking and an ambient intelligence display.

Proactive medication assistances based on spatiotemporal context awareness of aged persons.

In this paper, we developed a system that could assist appropriate activities for medication adherence of the elderly. It employs a proactive knowledge which is represented as templates predefined for their medication activities. The knowledge-based assistance depends on the contexts considerably, which the system can recognize by continuously monitoring the current position and the time-schedule for their medications. The monitoring is performed with ultrasonic sensors and infrared sensors mounted in a display and a pillbox mainly. According to the recognized contexts, the medication activities can he serviced through old person-friendly multimedia display. In special, since the knowledge is well-defined by XML, only its content modification can provide a variety of services individually customized to the elderly.

A semantic inheritance/inverse-inheritance mechanism for systematic bio-ontology construction.

In this paper, we propose a semantic inheritance/inverse-inheritance mechanism for systematic bio-ontology construction. This mechanism allows domain experts to easily manage sophisticated bio-ontologies in which biological knowledge is encoded; it automatically captures semantics inferred from the ontology structure being constructed or already constructed. Based on the captured semantics it suggests appropriate recommendation to the experts. While inheritance enables them to consistently determine the semantics of relationships between ontology concepts (or classes), inverse-inheritance allows them to incrementally refine the semantics by exploiting a huge amount of relationships between the instances of the concepts. To demonstrate the feasibility of the mechanism, we also implement an OWL(Web Ontology Language)-based graphical bio-ontology management system. In the system, the mechanism is seamlessly applied to the ontology by well defined graphic notations based on OWL. OWL is adopted to fully express the subtle semantics inherently buried in the bio-ontology.

A Hub-protein based visualization of large protein-protein interaction networks.

According as the protein-protein interaction (PPI) data more increase, we need to optimally visualize them as network, in that describing the relationship among proteins is able to easily analyze biological processes happened in a cell. In this paper, to fast layout large-scale PPI networks, we proposed a method taking hub-proteins into consideration, which have more interactions than any other proteins in a network. In other words, it enforces two core parts of Walshaw's multilevel force-directed placement algorithm (MLFDP) to be modified. The modification is achieved by coarsening and expanding all neighboring proteins of hub-protein just once, whereas only two proteins in Walshaw's method. Our experiments show that the quality of layout is better optimal and time cost is reduced up to 63% in comparison with other methods.

A rule-based detection of functional modules in protein-protein interaction networks.

In the protein-protein interaction (PPI) network there are many functional modules, each involving several protein interactions to perform discrete functions. Pathways and protein complexes are the examples of the functional modules. In this paper, we propose a rule-based method for detecting the modules. The rule is expressed in terms of triples and operators between the triples. The former represents conceptual relations reifying the protein interactions of a module, and the latter defines the structure of the module with the relations. Additionally, users can define composite rules by composing the predefined rules. The composite rules make it possible to detect modules that are conceptually similar as well as structurally identical to users' queries. The rules are managed in the XML format so that they can be easily applied to other networks of different species. We also provide a visualized environment for intuitionally describing complexly structured rules.