Holdout Cross-validation Research Articles

High-Spatial-Resolution Estimates of Ultrafine Particle Concentrations across the Continental United States.

There is growing evidence that ultrafine particles (UFP; particles smaller than 100 nm) are likely more toxic than larger particles. However, the health effects of UFP remain uncertain due in part to the lack of large-scale population-based exposure assessment. We develop a national-scale empirical model of particle number concentration (PNC; a measure of UFP) using data from mobile monitoring and fixed sites across the United States and a land-use regression (LUR) modeling framework. Traffic, commercial land use, and urbanicity-related variables explain much of the spatial variability of PNC (base model R2 = 0.77, RMSE = 2400 cm-3). Model predictions are robust across a diverse set of evaluations [random 10-fold holdout cross-validation (HCV): R2 = 0.72, RMSE = 2700 cm-3; spatially defined HCV: R2 = 0.66, RMSE = 3000 cm-3; evaluation against an independent data set: R2 = 0.54, RMSE = 2600 cm-3]. We apply our model to predict PNC at ∼6 million residential census blocks in the contiguous United States. Our estimates are annual average concentrations for 2016-2017. The predicted national census-block-level mean PNC ranges between 1800 and 26 600 cm-3 (population-weighted average: 6500 cm-3), with hotspots in cities and near highways. Our national PNC model predicts large urban-rural, intra-, and inter-city contrasts. PNC and PM2.5 are moderately correlated at the city scale, but uncorrelated at the regional/national scale. Our high-spatial-resolution national PNC estimates are useful for analyzing population exposure (socioeconomic disparity, epidemiological health impact) and environmental policy and regulation.

Machine learning enables noninvasive prediction of atrial fibrillation driver location and acute pulmonary vein ablation success using the 12-lead ECG

BackgroundAtrial fibrillation (AF) is the most common supraventricular arrhythmia, characterized by disorganized atrial electrical activity, maintained by localized arrhythmogenic atrial drivers. Pulmonary vein isolation (PVI) allows to exclude PV-related drivers. However, PVI is less effective in patients with additional extra-PV arrhythmogenic drivers.ObjectivesTo discriminate whether AF drivers are located near the PVs vs extra-PV regions using the noninvasive 12-lead electrocardiogram (ECG) in a computational and clinical framework, and to computationally predict the acute success of PVI in these cohorts of data.MethodsAF drivers were induced in 2 computerized atrial models and combined with 8 torso models, resulting in 1128 12-lead ECGs (80 ECGs with AF drivers located in the PVs and 1048 in extra-PV areas). A total of 103 features were extracted from the signals. Binary decision tree classifier was trained on the simulated data and evaluated using hold-out cross-validation. The PVs were subsequently isolated in the models to assess PVI success. Finally, the classifier was tested on a clinical dataset (46 patients: 23 PV-dependent AF and 23 with additional extra-PV sources).ResultsThe classifier yielded 82.6% specificity and 73.9% sensitivity for detecting PV drivers on the clinical data. Consistency analysis on the 46 patients resulted in 93.5% results match. Applying PVI on the simulated AF cases terminated AF in 100% of the cases in the PV class.ConclusionMachine learning–based classification of 12-lead-ECG allows discrimination between patients with PV drivers vs those with extra-PV drivers of AF. The novel algorithm may aid to identify patients with high acute success rates to PVI.

Comparison of gap-filling techniques applied to the CCI soil moisture database in Southern Europe

Soil moisture (SM) is a key variable that plays an important role in land-atmosphere interactions. Monitoring SM is crucial for many applications and can help to determine the impact of climate change. Therefore, it is essential to have continuous and long-term databases for this variable. Satellite missions have contributed to this; however, the continuity of the series is compromised due to the data gaps derived by different factors, including revisit time, presence of seasonal ice or Radio Frequency Interference (RFI) contamination. In this work, the applicability of different gap-filling techniques is evaluated on the ESA Climate Change Initiative (CCI) SM combined product, which is the longest available satellite-based SM data record. The methods used were linear, cubic and autoregressive interpolation and support vector machines (SVMs). This study focused on Southern Europe and spanned the years 2003–2015. The different methods were applied in the temporal and spatial domains and evaluated using the holdout cross-validation technique. A set of variables was introduced in the SVM model to estimate SM, namely, land surface temperature, precipitation, normalized difference vegetation index (NDVI), potential evaporation, soil texture and geographical coordinates. For the SVMs, several combinations of these variables were considered, including a principal component analysis (PCA) containing all of them. Although the different methods show a generally good performance, the SVM method outperforms the rest. Using the SM of the precedent day (SMt-1) is key to obtain good estimates. The median value of the correlation coefficient (R) obtained with the SVM and the SMt-1 series in the temporal analysis was 0.83, and the RMSE was 0.025 m3m−3. Similar results were obtained in the spatial analysis, with the best performance (R = 0.88; RMSE = 0.024 m3m−3) obtained by the SVM using the SMt-1 series and the static variables. The application of PCA to input variables was not beneficial, and the interpolation methods failed when dealing with large spatial or temporal gaps. A validation of the CCI SM series with in situ SM data from four networks located in Spain, France, Germany and Italy was also performed and no substantial differences were observed between results obtained with the original and with the reconstructed series. In addition, best inputs obtained with SVM were used to evaluate the random forest (RF) method in the temporal and spatial domain. This method showed a good ability to estimate soil moisture values in the temporal domain but to a lesser extent than SVM while for the spatial domain it did not seem to be as accurate. Our results confirm that we can efficiently deal with spatio-temporal gaps on observational SM databases using the SVM method and the past time series and soil texture as supporting information.

An Intelligent Business Model for Product Price Prediction Using Machine Learning Approach

The price of a product plays a vital role in its market share. Customers usually buy a product when it fits their needs and budget. Therefore, it is an essential area in the business to make decisions about prices for each product. The major portion of the business profit is directly connected with the percentage of the sale, which relies on certain factors of customers including customers’ behavior and market competitors. It has been observed in the past that machine learning algorithms have made the decision-making process more effective and profitable in businesses. The fusion of machine learning with business intelligence systems has produced potentially better results for business workers. This study proposes a fusion model of machine learning and business intelligence to facilitate the effective pricing mechanism of products. Three machine learning algorithms namely (1) Multiclass Random Forest, (2) Multiclass Logistic Regression and (3) Multiclass one-vs-all have been applied for product price prediction. This intelligent business model helps business workers to set product pricing and discounts based on customer’s buying behavior (i.e., customer’s average amount spent in buying items). The model also predicts the demand for the product in the market, which may help in making effective decisions about the product to gain a better productivity. The performance of the model has been evaluated using hold-out cross validation. Multiclass one-vs-all out performed the other two algorithms in terms of average log-loss and it produced better accuracy in prediction for both hold-out as well as cross validation. The outcomes of the proposed model may help businesses in exploring business data and to gain higher share in the market.

Majority voting-based hybrid feature selection in machine learning paradigm for epilepsy detection using EEG

This article presents a combination of statistical and discrete wavelet transform (DWT)-based features for the identification of epileptic seizures in electroencephalogram (EEG) signals. A total of 150 quantitative features are extracted from EEG signals. A multi-criteria hybrid feature selection is proposed by combining six feature ranking methods using the majority voting technique to identify the most relevant EEG markers. Kernel-based support vector machine is used to evaluate the proposed approach along with a hybrid classifier namely support vector neural network (SVNN) which is a combination of support vector machine (SVM) and artificial neural network (ANN). For performance evaluation of the proposed method, a benchmarked database is used. A comparative study of various types of SVM and SVNN with ten-fold and hold-out cross-validation techniques is conducted. The highest classification accuracy (CA) of 98.18% and 100% sensitivity is achieved with a fine Gaussian SVM classifier with hold-out data division protocol.

Majority voting-based hybrid feature selection in machine learning paradigm for epilepsy detection using EEG

This article presents a combination of statistical and discrete wavelet transform (DWT)-based features for the identification of epileptic seizures in electroencephalogram (EEG) signals. A total of 150 quantitative features are extracted from EEG signals. A multi-criteria hybrid feature selection is proposed by combining six feature ranking methods using the majority voting technique to identify the most relevant EEG markers. Kernel-based support vector machine is used to evaluate the proposed approach along with a hybrid classifier namely support vector neural network (SVNN) which is a combination of support vector machine (SVM) and artificial neural network (ANN). For performance evaluation of the proposed method, a benchmarked database is used. A comparative study of various types of SVM and SVNN with ten-fold and hold-out cross-validation techniques is conducted. The highest classification accuracy (CA) of 98.18% and 100% sensitivity is achieved with a fine Gaussian SVM classifier with hold-out data division protocol.

Deep Layer Kernel Sparse Representation Network for the Detection of Heart Valve Ailments from the Time-Frequency Representation of PCG Recordings.

The heart valve ailments (HVAs) are due to the defects in the valves of the heart and if untreated may cause heart failure, clots, and even sudden cardiac death. Automated early detection of HVAs is necessary in the hospitals for proper diagnosis of pathological cases, to provide timely treatment, and to reduce the mortality rate. The heart valve abnormalities will alter the heart sound and murmurs which can be faithfully captured by phonocardiogram (PCG) recordings. In this paper, a time-frequency based deep layer kernel sparse representation network (DLKSRN) is proposed for the detection of various HVAs using PCG signals. Spline kernel-based Chirplet transform (SCT) is used to evaluate the time-frequency representation of PCG recording, and the features like L1-norm (LN), sample entropy (SEN), and permutation entropy (PEN) are extracted from the different frequency components of the time-frequency representation of PCG recording. The DLKSRN formulated using the hidden layers of extreme learning machine- (ELM-) autoencoders and kernel sparse representation (KSR) is used for the classification of PCG recordings as normal, and pathology cases such as mitral valve prolapse (MVP), mitral regurgitation (MR), aortic stenosis (AS), and mitral stenosis (MS). The proposed approach has been evaluated using PCG recordings from both public and private databases, and the results demonstrated that an average sensitivity of 100%, 97.51%, 99.00%, 98.72%, and 99.13% are obtained for normal, MVP, MR, AS, and MS cases using the hold-out cross-validation (CV) method. The proposed approach is applicable for the Internet of Things- (IoT-) driven smart healthcare system for the accurate detection of HVAs.

PSXI-22 Prediction quality of cattle behavior traits evaluated through different cross-validation strategies using wearable sensor data and machine learning algorithms

Abstract Wearable sensors have been adopted as an alternative for real-time monitoring of cattle feeding behavior in grazing systems. However, even using machine learning (ML) techniques confounding effects such as cross-validation strategy may inflate the prediction quality. Our objective was to evaluate the effect of different cross-validation strategies on the prediction of grazing activities in cattle using wearable sensor data and ML algorithms. Six Nellore bulls (345 ± 21 kg) had their behavior visually classified as grazing or not-grazing for a period of 15 days. Generalized Linear Model (GLM), Random Forest (RF), and Artificial Neural Network (ANN) were employed to predict behavior (grazing or not-grazing) using 3-axis accelerometer data. For each analytical method, three cross-validation strategies were evaluated: holdout, leave-one-animal-out (LOAO), and leave-one-day-out (LODO). Algorithms were trained using similar dataset sizes (holdout: n = 57,862; LOAO: n = 56,786; LODO: n = 56,672). Regardless of the cross-validation strategy, GLM achieved the worst prediction accuracy (53%) compared to the ML techniques (65% for both RF and ANN). ANN performed slightly better than RF for LOAO (73%) and LODO (64%) cross-validation strategies. The holdout yielded the highest accuracy values for all three ML approaches (GLM: 59%, RF: 76%, and ANN: 74%), followed by LODO (58%) and LOAO (55%). In conclusion, the GLM approach was not adequate to predict grazing behavior, regardless of the cross-validation strategy. The greater prediction accuracy observed for holdout cross-validation may simply indicate a lack of data independence and the presence of carry-over effects from animals and grazing management. Our results suggest that generalizing predictive models to unknown (not used for training) animals or grazing management may incur in poor prediction quality. The results highlight the need for using biological knowledge to define the validation strategy that is closer to the real-life situation.

Abstract 14955: Machine Learning Predicts Hemodynamic Instability in Children After Cardiac Surgery in Pediatric Intensive Care Unit (PICU)

Introduction: Despite improvements in management for children after cardiac surgery, a non-negligible proportion of patients suffer from cardiac arrest, having a poor prognosis. Although serum lactate levels are widely accepted markers of hemodynamic instability, measuring lactate requires discrete blood sampling. An alternative method to evaluate hemodynamic stability/instability continuously and non-invasively may assist in improving the standard of patient care. Hypothesis: We hypothesize that blood lactate in PICU patients can be predicted using machine learning applied to arterial waveforms and perioperative characteristics. Methods: Forty-eight children, who underwent heart surgery, were included. Patient characteristics and physiological measurements were acquired and analyzed using specialized software/hardware, including heart rate, lactate level, arterial waveform sharpness, and area under the curve. Predicting a patient’s blood lactate levels was accomplished using regression-based supervised learning algorithms, including regression decision trees, tuned decision trees, random forest regressor, tuned random forest, AdaBoost regressor, and hypertuned AdaBoost. All algorithms were compared with hold-out cross validation. Two approaches were considered: basing prediction on the currently acquired physiological measurements along with those acquired at admission, as well as adding the most recent lactate measurement and the time since that measurement as prediction parameters. The second approach supports updating the learning system’s predictive capacity whenever a patient has a new ground truth blood lactate reading acquired. Results: In both approaches, the best performing machine learning method was the tuned random forest, which yielded a mean absolute error of 5.60 mg/dL in the first approach, and 4.62 mg/dL when predicting blood lactate with updated ground truth. Conclusions: In conclusion, the tuned random forest is capable of predicting the level of serum lactate by analyzing perioperative variables, including the arterial pressure waveform. Machine learning can predict the patient’s hemodynamics non-invasively, continuously, and with accuracy that may demonstrate clinical utility.

Modeling and analysis of a time series of equine infectious anemia cases in the state of Tocantins, Brazil, between 2007 and 2019

Equine infectious anemia (EIA) is a viral infectious disease that affects Equidae and is clinically characterized by intermittent fever, anemia, depression, emaciation, and edema. To evaluate disease dynamics in the state of Tocantins, Brazil, a time series of EIA cases in the period 2007–2019 was analyzed to describe the pattern of occurrence and define the autoregressive integrated by moving average (ARIMA) model best suited to make predictions of cases of this disease for the period 2020–2021. The modeling and statistical analysis of the time series were performed using R software. The ARIMA model (2,1,1) was evaluated by Holdout cross-validation, in which data from the periods 2007–2017 and 2018–2019 were used as training and test data, respectively. The analyses showed that EIA was endemic and non-seasonal in Tocantins. The ARIMA model (2,1,1) showed good predictive capacity adjusted for this time series. However, the prediction of 276 cases of EIA in Tocantins for the period 2020–2021 may vary depending on the demand for diagnostic tests for Equidae transportation and herd sanitation in farms considered infection foci. The ARIMA model helps predict the number of EIA cases in Tocantins and improves planning for disease control by the Official Veterinary Service.

Evaluating Lumbar Shape Deformation With Fabric Strain Sensors

To better study human motion inside the space suit and suit-related contact, a multifactor statistical model was developed to predict torso body shape changes and lumbar motion during suited movement by using fabric strain sensors that are placed on the body. Physical interactions within pressurized space suits can pose an injury risk for astronauts during extravehicular activity (EVA). In particular, poor suit fit can result in an injury due to reduced performance capabilities and excessive body contact within the suit during movement. A wearable solution is needed to measure body motion inside the space suit. An array of flexible strain sensors was attached to the body of 12 male study participants. The participants performed specific static lumbar postures while 3D body scans and sensor measurements were collected. A model was created to predict the body shape as a function of sensor signal and the accuracy was evaluated using holdout cross-validation. Predictions from the torso shape model had an average root mean square error (RMSE) of 2.02 cm. Subtle soft tissue deformations such as skin folding and bulges were accurately replicated in the shape prediction. Differences in posture type did not affect the prediction error. This method provides a useful tool for suited testing and the information gained will drive the development of injury countermeasures and improve suit fit assessments. In addition to space suit design applications, this technique can provide a lightweight and wearable system to perform ergonomic evaluations in field assessments.

Forecasting maximum surface settlement caused by urban tunneling

In this article, maximum surface settlement (MSS) of urban tunnels was investigated on the basis of three operational parameters of tunnel width, tunnel depth, excavation method, as well as three soil parameters of cohesion, friction angle and elasticity modulus. Seven intelligent methods of long short-term memory (LSTM), deep neural networks (DNNs), K-nearest neighbor (KNN), Gaussian process regression (GPR), support vector regression (SVR), decision tree (DT), and linear regression (LR) were used to perform investigation. The intelligent methods were studied on the basis of 300 datasets accessed from 8 urban tunnels in Iran. Two cross-validation methods of hold-out and 5-fold were utilized for analyzing the prediction results. Finally, the DNNs method with R2 = 0.9939 and RMSE = 3.396301689 mm in the hold-out cross-validation mode and R2 = 0.9937 and RMSE = 2.199337605 mm in the 5-fold cross-validation mode, was recommended and suggested as the best prediction method for MSS.

Development and validation of a sample entropy-based method to identify complex patient-ventilator interactions during mechanical ventilation

Patient-ventilator asynchronies can be detected by close monitoring of ventilator screens by clinicians or through automated algorithms. However, detecting complex patient-ventilator interactions (CP-VI), consisting of changes in the respiratory rate and/or clusters of asynchronies, is a challenge. Sample Entropy (SE) of airway flow (SE-Flow) and airway pressure (SE-Paw) waveforms obtained from 27 critically ill patients was used to develop and validate an automated algorithm for detecting CP-VI. The algorithm’s performance was compared versus the gold standard (the ventilator’s waveform recordings for CP-VI were scored visually by three experts; Fleiss’ kappa = 0.90 (0.87–0.93)). A repeated holdout cross-validation procedure using the Matthews correlation coefficient (MCC) as a measure of effectiveness was used for optimization of different combinations of SE settings (embedding dimension, m, and tolerance value, r), derived SE features (mean and maximum values), and the thresholds of change (Th) from patient’s own baseline SE value. The most accurate results were obtained using the maximum values of SE-Flow (m = 2, r = 0.2, Th = 25%) and SE-Paw (m = 4, r = 0.2, Th = 30%) which report MCCs of 0.85 (0.78–0.86) and 0.78 (0.78–0.85), and accuracies of 0.93 (0.89–0.93) and 0.89 (0.89–0.93), respectively. This approach promises an improvement in the accurate detection of CP-VI, and future study of their clinical implications.

Prediction equation for maximum stress of concrete drainage pipelines subjected to various damages and complex service conditions

Concrete drainage pipelines built in early China are seriously aged and in disrepair. Therefore, the methods that can be directly used to evaluate the service performances of such pipelines are urgently needed. In this study, 100 three-dimensional (3-D) refined finite element (FE) models of concrete drainage pipelines subjected to corrosion and void damages were established based on ABAQUS 6.14-1/Standard. First, the full-scale test data in the existing studies were used to verify the FE model. Then, 100 numerical analyses were performed on 9 dimensionless variables affecting the stress of the pipelines, among them, 70% were randomly selected as the training set, and 30% as the test set based on the Holdout cross-validation method. Finally, the maximum stress evaluation model for the concrete drainage pipelines subjected to damages and complex service conditions was successfully proposed based on multivariate nonlinear regression (MNR). The training and test results show that the proposed equation is capable of predicting the maximum stress of the pipeline with high accuracy (R2 = 0.96 and 0.97, respectively), and the prediction error is in normal distribution. Verification of the proposed equation shows that it has a high accuracy in predicting the effects of h, Ch, Cl, and Vl on the maximum stress of the pipelines, with the relative errors of 13.55%, 12.70%, 13.33%, and 18.94%, respectively. In particular, for traffic load, the relative error is only 4.97%. Additionally, the Spearman’s rank correlation coefficient of each variable was calculated and normalized to analyze the relative contribution of each variable to the pipeline maximum stress. The analyses show that the traffic load has the greatest effect on the pipeline stress, with a relative contribution percentage of 25.1%. Furthermore, the feasibility of the proposed equation in predicting the remaining life of pipelines is proved.

Breast lesion characterization using Quantitative Ultrasound (QUS) and derivative texture methods.

PurposeAccurate and timely diagnosis of breast cancer is extremely important because of its high incidence and high morbidity. Early diagnosis of breast cancer through screening can improve overall prognosis. Currently, biopsy remains as the gold standard for tumor pathological confirmation. Development of diagnostic imaging techniques for rapid and accurate characterization of breast lesions is required. We aim to evaluate the usefulness of texture-derivate features of QUS spectral parametric images for non-invasive characterization of breast lesions.MethodsQUS Spectroscopy was used to determine parametric images of mid-band fit (MBF), spectral slope (SS), spectral intercept (SI), average scatterer diameter (ASD), and average acoustic concentration (AAC) in 204 patients with suspicious breast lesions. Subsequently, texture analysis techniques were used to generate texture maps from parametric images to quantify heterogeneities of QUS parametric images. Further, a second-pass texture analysis was applied to obtain texture-derivate features. QUS parameters, texture-parameters and texture-derivate parameters were determined from both tumor core and a 5-mm tumor margin and were used in comparison to histopathological analysis in order to develop a diagnostic model for classifying breast lesions as either benign or malignant. Both leave-one-out and hold-out cross-validations were used to evaluate the performance of the diagnostic model. Three standard classification algorithms including a linear discriminant analysis (LDA), k-nearest neighbors (KNN), and support vector machines-radial basis function (SVM-RBF) were evaluated.ResultsCore and margin information using the SVM-RBF attained the best classification performance of 90% sensitivity, 92% specificity, 91% accuracy, and 0.93 AUC utilizing QUS parameters and their texture derivatives, evaluated using leave-one-out cross-validation. Implementation of hold-out cross-validation using combination of both core and margin information and SVM-RBF achieved average accuracy and AUC of 88% and 0.92, respectively.ConclusionsQUS-based framework and derivative texture methods enable accurate classification of breast lesions. Evaluation of the proposed technique on a large cohort using hold-out cross-validation demonstrates its robustness and its generalization.

Lateralized Connectivity between Globus Pallidus and Motor Cortex is Associated with Freezing of Gait in Parkinson’s Disease

Freezing of gait (FoG) is a brief, episodic absence or marked reduction of forward progression of the feet, despite the intention to walk, that is common in people with Parkinson’s disease (PD). We hypothesized that not only motor, but higher level cognitive and attention areas may be impaired in freezers. In this study, we aimed to characterize differences in cortical and subcortical functional connectivity specific to FoG. We examined resting state neuroimaging and objective measures of FoG severity and gait from 103 individuals (28 PD + FoG, 36 PD − FoG and 39 healthy controls). Inertial sensors were used to quantify freezing severity and gait. Groups with and without FoG were matched on age, disease severity, cognitive status, and levodopa medication. MRI data was processed using surface-based registration. High-quality imaging data were used to characterize differences in connectivity specific to FoG using a pre-defined set of Regions of Interest (ROIs) and validated using whole-brain connectivity analysis. Associations between functional connectivity and objective measures of FoG were determined via predictive modeling using hold-out cross validation. We found that connectivity between the left globus pallidus (GP) and left somatosensory cortex and between two brain areas in the default and insular/vestibular networks exhibited significant differences specific to FoG and were also strong and significant predictors of FoG severity. Our findings suggest that the interplay among motor, default and vestibular areas of the left cortex are critical in the pathology of FoG.

Accuracy Of Classifying Prediabetes Predicted By Grip Strength In Obese Adults Utilizing Machine Learning

PURPOSE: To compare the accuracy of classifying prediabetes, include type 2 diabetes, predicted by hand-grip strength employing machine learning (ML) techniques in Korean obese adults. METHODS: The data of 1230 Korean obese adults (51.5% males, 19-65yrs) was retrieved from the Korean National Health and Nutrition Examination Survey (KNHANES) 2014-2015. Obesity was identified by the Korean standard BMI (BMI ≥ 25kg/m2) and waist circumference (WC; Male: WC ≥ 90cm, Female: WC ≥ 85cm). A total of 591 individuals with prediabetes and type 2 diabetes was diagnosed by the criteria 1) diagnosed as type 2 diabetes, 2) using diabetes medication, 3) abnormal fasting glucose level (fasting glucose ≥ 100mg/dL). Three grip strength models were employed, which were 1) normal grip strength (GS), 2) divided by weight (GSW), and 3) divided by BMI (GSB), respectively, to examine the effect of different relative grip strengths on prediabetes. Multilayer perceptron (MLP) and traditional logistic regression (LR) algorithms employing RSNNS package in R, were applied to classify the prediabetes predicted by hand-grip adjusted by age, income, education, occupation, marital status, binge drinking, smoking, daily calories intake, sedentary time, strengthening exercise adherence, aerobic exercise adherence variables, separated by gender. To evaluate the accuracy (ACC), sensitivity (SEN), and specificity (SPE) of the confusion matrix of ML, the participants were separated into a train for deriving equations and a test group for holdout cross-validation. RESULTS: To classify prediabetes in obese males, GSB adjusted by co-variates was revealed the highest ACC for both ML classifiers (train group: MLP=71.5%, LR=67.6%; test group: MLP=63.9%, LR=66.1%), and GSW showed the lowest ACC (MLP: train=67.6%, test=61.6%). Moreover, GSW and GSB of MLP showed higher SEN than SPE in test group (SEN, SPE: GSW=78.6%, 44.1%; GSB=74.4%, 52.5%, respectively). In obese females, however, both ML classifiers did not show any consistent accuracy levels. CONCLUSIONS: It was revealed that GSB showed a strong relationship with prediabetes in obese male adults in Korea. Moreover, MLP and LR classifiers present fair accuracy in the cross-validation to classify prediabetes predicted by GSB.

In Silico Guidance for In Vitro Androgen and Glucocorticoid Receptor ToxCast Assays.

Molecular initiating events (MIEs) are key events in adverse outcome pathways that link molecular chemistry to target biology. As they are based on chemistry, these interactions are excellent targets for computational chemistry approaches to in silico modeling. In this work, we aim to link ligand chemical structures to MIEs for androgen receptor (AR) and glucocorticoid receptor (GR) binding using ToxCast data. This has been done using an automated computational algorithm to perform maximal common substructure searches on chemical binders for each target from the ToxCast dataset. The models developed show a high level of accuracy, correctly assigning 87.20% of AR binders and 96.81% of GR binders in a 25% test set using holdout cross-validation. The 2D structural alerts developed can be used as in silico models to predict these MIEs and as guidance for in vitro ToxCast assays to confirm hits. These models can target such experimental work, reducing the number of assays to be performed to gain required toxicological insight. Development of these models has also allowed some structural alerts to be identified as predictors for agonist or antagonist behavior at the receptor target. This work represents a first step in using computational methods to guide and target experimental approaches.

Development and evaluation of the population pharmacokinetic models for FVIII and FIX concentrates of the WAPPS-Hemo project.

The Web-Accessible Population Pharmacokinetic Service (WAPPS) project generates individually predicted pharmacokinetic (PK) profiles and tailored prophylactic treatment regimens for haemophilic patients, which rely on a set of population PK (PopPK) models providing concentrate-specific priors for the Bayesian forecasting methodology. To describe the predictive performance of the WAPPS PopPK models in use on the WAPPS-Hemo platform. Data for modelling include dense PK data obtained from industry sponsored and independent PK studies, and dense and sparse data accumulated through WAPPS-Hemo. WAPPS PopPK models were developed via non-linear mixed-effect modelling taking into account the effects of covariates and between-individual-and sometimes between-occasion-variability. Model evaluation consisted of (a) prediction-corrected Visual Predictive Check (pcVPC), (b) Limited Sampling Analysis (LSA) and (c) repeated hold-out cross-validation. Thirty-three WAPPS PopPK models built on data from 3188 patients (ages 1-78years) under treatment by factor VIII or IX products (FVIII, FIX) were evaluated. Overall, models exhibit excellent performance characteristics. The pcVPC shows that the observed PK data fall within acceptable 90% interpercentile predictive bands. A slight overprediction beyond the expected half-life, an anticipated result of using sparse data, occurs for some models. The LSA results in lower than 3% of relative error for FVIII and FIX products and 16% for engineered FIX products. Cross-Validation analysis yields relative errors lower than 1.5% and 1.4% in estimates of half-life and time to 0.02IU/mL, respectively. The WAPPS-Hemo models consistently showed excellent performance characteristics for the intended use for Bayesian forecasting of individual PK profiles.

Transdiagnostic Brain Mapping in Developmental Disorders.

SummaryChildhood learning difficulties and developmental disorders are common, but progress toward understanding their underlying brain mechanisms has been slow. Structural neuroimaging, cognitive, and learning data were collected from 479 children (299 boys, ranging in age from 62 to 223 months), 337 of whom had been referred to the study on the basis of learning-related cognitive problems. Machine learning identified different cognitive profiles within the sample, and hold-out cross-validation showed that these profiles were significantly associated with children’s learning ability. The same machine learning approach was applied to cortical morphology data to identify different brain profiles. Hold-out cross-validation demonstrated that these were significantly associated with children’s cognitive profiles. Crucially, these mappings were not one-to-one. The same neural profile could be associated with different cognitive impairments across different children. One possibility is that the organization of some children’s brains is less susceptible to local deficits. This was tested by using diffusion-weighted imaging (DWI) to construct whole-brain white-matter connectomes. A simulated attack on each child’s connectome revealed that some brain networks were strongly organized around highly connected hubs. Children with these networks had only selective cognitive impairments or no cognitive impairments at all. By contrast, the same attacks had a significantly different impact on some children’s networks, because their brain efficiency was less critically dependent on hubs. These children had the most widespread and severe cognitive impairments. On this basis, we propose a new framework in which the nature and mechanisms of brain-to-cognition relationships are moderated by the organizational context of the overall network.