Cross-validated R2 Research Articles

Multiple bioanalytical method based residual biomass prediction in microbial culture using multivariate regression and artificial neural network

Optical density (OD) based measurements cannot be a reliable proxy for residual biomass concentration in polyhydroxyalkanoate (PHA) accumulating microbial culture, as cell size and morphology change during growth. In this study, four independent analytical methods i.e., OD measurement at 540 nm and 600 nm, tetrazolium reduction assay, and intracellular protein estimation were adopted to model residual biomass growth in Cupriavidus necator. The inter-day variation of calibration slope for residual biomass was significant (p < 0.001), and the regression coefficient (R2) of composite samples across the methods varied between 0.74 and 0.96. A reduced quadratic polynomial model (R2, 0.996; adjusted R2, 0.995, cross-validation R2 0.987) was chosen to predict residual biomass using multi-analytical measurements. Partial least square regression and variable selection suggested the inclusion of OD 540 nm and protein measurement into two retained latent variables, with R2 of 0.986, and adjusted R2 of 0.972. ANN model offered good predictability for residual biomass, showing close agreement of experimental and modelled datasets for training, validation, and test subsets with R2 of 0.998, 0.994, and 0.943, respectively. Analytical sensitivity of quadratic and PLS regression models were 0.058 and 0.164 respectively. A comparison of the proposed methods suggests that all three can be used as an alternative to dry-cell-weight-based residual biomass estimation, avoiding lengthy drying time and PHA extraction and measurement steps.

Neighborhood-scale ambient NO2 concentrations using TROPOMI NO2 data: Applications for spatially comprehensive exposure assessment

This study estimated long-term average ambient NO2 concentrations using TROPOspheric Monitoring Instrument (TROPOMI) tropospheric NO2 data and land use information at the spatial resolution of 500 m in California for the years 2018–2019. Our satellite-land use regression model demonstrated reasonably high predictive power with cross-validation (CV) R2 = 0.76, mean absolute error (MAE) = 1.95 ppb, and root mean squared error (RMSE) = 2.51 ppb in a comparison between measured and estimated NO2 concentrations. Exploiting the high-resolution NO2 estimates, we further investigated the representativeness of ground NO2 monitors for population exposures and examined the spatial variation of NO2 in relation to parcel-level property data for exposure attributions. The ground NO2 monitors were the most representative of population exposures in Los Angeles and San Diego counties, supported by population-weighted average NO2 concentrations (satellite-derived estimations) similar to arithmetic average NO2 concentrations (ground measurements). On the contrary, the exposure assessment using the ground monitors was the least representative and protective in Humboldt, San Luis Obispo, and Yolo counties with population-weighted average NO2 greater than arithmetic average NO2 by 82.2 % (1.85 ppb), 67.1 % (1.89 ppb), and 58.2 % (2.48 ppb), respectively. In a case study of LA County, we identified comparatively high NO2 concentrations for the property types of food processing facilities and high-density residential complexes (such as high-rise apartments and apartments). This finding provides evidence that these emerging sources may be crucial to mitigate cumulative NO2 exposures and subsequent health risks from a regulatory perspective.

Estimation of surface ozone concentration over Jiangsu province using a high-performance deep learning model

Recently, the global background concentration of ozone (O3) has demonstrated a rising trend. Among various methods, groun-based monitoring of O3 concentrations is highly reliable for research analysis. To obtain information on the spatial characteristics of O3 concentrations, it is necessary that the ground monitoring sites be constructed in sufficient density. In recent years, many researchers have used machine learning models to estimate surface O3 concentrations, which cannot fully provide the spatial and temporal information contained in a sample dataset. To solve this problem, the current study utilized a deep learning model called the Residual connection Convolutional Long Short-Term Memory network (R-ConvLSTM) to estimate daily maximum 8-hr average (MDA8) O3 over Jiangsu province, China during 2020. In this research, the R-ConvLSTM model not only provides the spatiotemporal information of MDA8 O3, but also involves residual connection to avoid the problem of gradient explosion and gradient disappearance with the deepening of network layers. We utilized the TROPOMI total O3 column retrieved from Sentinel-5 Precursor, ERA5 reanalysis meteorological data, and other supplementary data to build a pre-trained dataset. The R-ConvLSTM model achieved an overall sample-base cross-validation (CV) R2 of 0.955 with root mean square error (RMSE) of 9.372 µg/m3. Model estimation also showed a city-based CV R2 of 0.896 with RMSE of 14.029 µg/m3, the highest MDA8 O3 in spring being 122.60 ± 31.60 µg/m3 and the lowest in winter being 69.93 ± 18.48 µg/m3.

Estimation of fine particulate matter in an arid area from visibility based on machine learning.

The absence of air pollution monitoring networks makes it difficult to assess historical fine particulate matter (PM2.5) exposures for countries in the areas, such as Kuwait, which are severe impacted by desert dust and anthropogenic pollution. We constructed an ensemble machine learning model to predict daily PM2.5 concentrations for regions lack of PM2.5 observations. The model was constructed based on daily PM2.5, visibility, and other meteorological data collected at two sites in Kuwait. Then, our model was applied to predict the daily level of PM2.5 concentrations for eight airports located in Kuwait and Iraq from 2013 to 2020. As compared to traditional statistic models, the proposed machine learning methods improved the accuracy in using visibility to predict daily PM2.5 concentrations with a cross-validation R2 of 0.68. The predicted level of daily PM2.5 concentrations were consistent with previous measurements. The predicted average yearly PM2.5 concentration for the eight stations is 50.65 µg/m3. For all stations, the monthly average PM2.5 concentrations reached their maximum in July and their minimum in November. These findings make it possible to retrospectively estimate daily PM2.5 exposures using the large-scale databases of historical visibility in regions with few particulate matter monitoring stations. The scarcity of air pollution ground monitoring networks makes it difficult to assess historical fine particulate matter exposures for countries in arid areas such as Kuwait. Visibility is closely related to atmospheric particulate matter concentrations and historical airport visibility records are commonly available in most countries. Our model make it possible to retrospectively estimate daily PM2.5 exposures using the large-scale databases of historical visibility in arid regions with few particulate matter ground monitoring stations. The product of such models can be critical for environmental risk assessments and population health studies.

Within-city variation in ambient carbon monoxide concentrations: Leveraging low-cost monitors in a spatiotemporal modeling framework

Exposure to ambient carbon monoxide (CO) may be associated with cardiovascular disease outcomes based on human and animal experimental evidence, but current epidemiologic research is limited. The number and distribution of ground-level regulatory agency monitors are insufficient to characterize potential fine-scale variation of CO. This study aimed to develop a high-resolution ambient CO prediction model at the daily scale based on both regulatory agency monitoring data and measurements from calibrated low-cost gas monitors in Baltimore, Maryland. We also evaluated the contribution of three novel parameters to model performance, including high-resolution meteorological data, satellite remote sensing data, and co-pollutant concentrations (PM2.5, NO2, and NOx). The CO model had spatial cross-validation (CV) R² (RMSE) of 0.70 (0.02 parts per million, ppm) and temporal CV R² (RMSE) of 0.61 (0.04 ppm). The predictions revealed spatially resolved CO hotspots associated with population, traffic, and other non-road emission sources (e.g., railroads and airport), as well as sharp concentration decreases within short distances from primary roads. The three novel parameters did not substantially improve model performance, suggesting that our spatiotemporal modeling framework based on geographic features was reliable and robust. As low-cost monitors become increasingly available, this approach to CO concentration modeling can be generalized to resource-restricted environments to facilitate comprehensive epidemiologic research.

Application of geostationary satellite and high-resolution meteorology data in estimating hourly PM2.5 levels during the Camp Fire episode in California

BACKGROUND: Wildland fire smoke contains large amounts of PM2.5 that can traverse tens to hundreds of kilometers, resulting in significant deterioration of air quality and excess mortality and morbidity in downwind regions. Estimating PM2.5 levels while considering the impact of wildfire smoke has been challenging due to the lack of ground monitoring coverage near the smoke plumes. AIM: We aim to estimate total PM2.5 concentration during the Camp Fire episode, the deadliest wildland fire in California history. METHODS: Our random forest (RF) model combines calibrated low-cost sensor data (PurpleAir) with regulatory monitor measurements (Air Quality System, AQS) to bolster ground observations, Geostationary Operational Environmental Satellite-16 (GOES-16)’s high temporal resolution to achieve hourly predictions, and oversampling techniques (Synthetic Minority Oversampling Technique, SMOTE) to reduce model underestimation at high PM2.5 levels. In addition, meteorological fields at 3 km resolution from the High-Resolution Rapid Refresh model and land use variables were also included in the model. RESULTS: Our AQS-only model achieved an out of bag (OOB) R2 (RMSE) of 0.84 (12.00 &amp;#x3bc;g/m3) and spatial and temporal cross-validation (CV) R2 (RMSE) of 0.74 (16.28 &amp;#x3bc;g/m3) and 0.73 (16.58 &amp;#x3bc;g/m3), respectively. Our AQS + Weighted PurpleAir Model achieved OOB R2 (RMSE) of 0.86 (9.52 &amp;#x3bc;g/m3) and spatial and temporal CV R2 (RMSE) of 0.75 (14.93 &amp;#x3bc;g/m3) and 0.79 (11.89 &amp;#x3bc;g/m3), respectively. Our AQS + Weighted PurpleAir + SMOTE Model achieved OOB R2 (RMSE) of 0.92 (10.44 &amp;#x3bc;g/m3) and spatial and temporal CV R2 (RMSE) of 0.84 (12.36 &amp;#x3bc;g/m3) and 0.85 (14.88 &amp;#x3bc;g/m3), respectively. CONCLUSIONS: Hourly predictions from our model may aid in epidemiological investigations of intense and acute exposure to PM2.5 during the Camp Fire episode. KEYWORDS: PM2.5, GOES16, Wildland fire, Remote sensing, AOD, SMOTE, Weighted Random Forest

Predicting fine-scale daily NO₂ over Mexico City using an ensemble modeling approach

Background and Aim: In recent years, there has been growing interest in developing air pollution prediction models to reduce exposure measurement error in epidemiologic studies. However, efforts for localized, fine-scale prediction models have been predominantly focused in the United States and Europe, with substantially less work conducted in low- and middle-income countries. Furthermore, the availability of new satellite instruments such as the TROPOspheric Monitoring Instrument (TROPOMI) provides novel opportunities for modeling efforts. Methods: We estimate daily ground-level nitrogen dioxide (NO₂) concentrations in the Mexico City Metropolitan Area at 1-km² grids from 2005 to 2019 using a multi-stage approach. In stage 1, we impute missing satellite NO₂ column measurements from the Ozone Monitoring Instrument (OMI) and TROPOMI using the random forest (RF) approach based on estimates from atmospheric ensemble models. In stage 2, we calibrate the association of column NO₂ to ground-level NO₂ using ground monitors and meteorological features using linear mixed-effects (LME) and RF models. In stage 3, we predict the stage 2 model over each 1-km² grid in our study area, then ensemble the results using a generalized additive model (GAM). In stage 4, we used RF to model the local component at the 200-m² scale by explaining the residual between predicted NO₂ from stage 3 and measured NO₂ at the monitoring stations. Results: The cross-validated R² of the LME and RF models in stage 2 were 0.69 and 0.75 respectively, and 0.83 for the ensembled GAM. Cross-validated root-mean-squared prediction error (RMSPE) of the GAM was 4.41 µg/m³. The RF model in stage 4 further explained on average 54% of the variation in the residual NO₂ concentrations. Conclusions: Using novel approaches and newly available remote sensing data, our multi-stage model presented high cross-validated fits and reconstructs fine-scale NO₂ estimates for further epidemiologic studies in Mexico City.

Early flowering changes robusta coffee yield responses to climate stress and management

A shift towards earlier flowering is a widely noted consequence of climate change for the world's plants. However, whether early flowering changes the way in which plants respond to climate stress, and in turn plant yield, remains largely unexplored. Using 10 years of flowering time and yield observations (Total N = 5580) from 558 robusta coffee (Coffea canephora) farms across Vietnam we used structural equation modelling (SEM) to examine the drivers of flowering day anomalies and the consequent effects of this on coffee climate stress sensitivity and management responses (i.e. irrigation and fertilization). SEM allowed us to model the cascading and interacting effects of differences in flowering time, growing season length and climate stress. Warm nights were the main driver of early flowering (i.e. flowering day anomalies <0), which in turn corresponded to longer growing seasons. Early flowering was linked to greater sensitivity of yield to temperature during flowering (i.e. early in the season). In contrast, when late flowering occurred yield was most sensitive to temperature and rainfall later in the growing season, after flowering and fruit development. The positive effects of tree age and fertilizer on yield, apparent under late flowering conditions, were absent when flowering occurred early. Late flowering models predicted yields under early flowering conditions poorly (a 50 % reduction in cross-validated R2 of 0.54 to 0.27). Likewise, models based on early flowering were unable to predict yields well under late flowering conditions (a 75 % reduction in cross-validated R2, from 0.58 to 0.14). Our results show that early flowering changes the sensitivity of coffee production to climate stress and management and in turn our ability to predict yield. Our results indicate that changes in plant phenology need to be taken into account in order to more accurately assess climate risk and management impacts on plant performance and crop yield.

A new comprehensive index for monitoring maize lodging severity using UAV-based multi-spectral imagery

Lodging significantly reduces crop yield and grain quality. Timely and accurately monitoring crop lodging severity to help farmers to adjust the field management and settle reasonable insurance claim. The purpose of this paper is to effectively monitor maize lodging severity using unmanned aerial vehicle (UAV) with a multi-spectral camera. Considering the plant height (PH), lodging angle (LA) and SPAD, each agronomic trait was transformed into a membership matrix by giving each factor a weight. The comprehensive superiority of each evaluation result based on agronomic trait (PH, LA, SPAD) was calculated by using the fuzzy comprehensive evaluation (FCE) method, and the comprehensive lodging evaluation index (CLEI) was constructed. The relationships between the spectral reflectance (SR), texture feature (TF), vegetation index (VI) and PH, LA, SPAD, CLEI were established. The CLEI was calculated and mapped by using the sensitive features derived from UAV-based multi-spectral imagery. According to the distribution of CLEI, the research area was divided into five grades, including non-lodging (NL), light lodging (LL), moderate lodging (ML), severe lodging (SL) and very severe lodging (VSL). The results showed that among the CLEI models constructed by single feature, VI performed best and the R2 of cross-validation was 0.66 which was higher than agronomic trait models (the R2 of PH, LA, and SPAD models were 0.56, 0.58, 0.41, respectively). The R2 and nRMSE of cross-validation of the CLEI model constructed by the top three sensitive features (CIre, NDRE, Re_Mea) were 0.80 and 23.78%, while the R2 of the PH, LA, and SPAD models were 0.74, 0.71, 0.46, respectively. The accuracy of the new comprehensive index (CLEI) model was better than that of single agronomic trait (PH, LA and SPAD) model. The CLEI proposed in this paper using UAV-based multi-spectral imagery can effectively monitor the lodging severity of maize.

National Land Use Regression Model for NO2 Using Street View Imagery and Satellite Observations.

Land use regression (LUR) models are widely applied to estimate intra-urban air pollution concentrations. National-scale LURs typically employ predictors from multiple curated geodatabases at neighborhood scales. In this study, we instead developed national NO2 models relying on innovative street-level predictors extracted from Google Street View [GSV] imagery. Using machine learning (random forest), we developed two types of models: (1) GSV-only models, which use only GSV features, and (2) GSV + OMI models, which also include satellite observations of NO2. Our results suggest that street view imagery alone may provide sufficient information to explain NO2 variation. Satellite observations can improve model performance, but the contribution decreases as more images are available. Random 10-fold cross-validation R2 of our best models were 0.88 (GSV-only) and 0.91 (GSV + OMI)─a performance that is comparable to traditional LUR approaches. Importantly, our models show that street-level features might have the potential to better capture intra-urban variation of NO2 pollution than traditional LUR. Collectively, our findings indicate that street view image-based modeling has great potential for building large-scale air quality models under a unified framework. Toward that goal, we describe a cost-effective image sampling strategy for future studies based on a systematic evaluation of image availability and model performance.

Within-City Variation in Ambient Carbon Monoxide Concentrations: Leveraging Low-Cost Monitors in a Spatiotemporal Modeling Framework.

Background:Based on human and animal experimental studies, exposure to ambient carbon monoxide (CO) may be associated with cardiovascular disease outcomes, but epidemiological evidence of this link is limited. The number and distribution of ground-level regulatory agency monitors are insufficient to characterize fine-scale variations in CO concentrations.Objectives:To develop a daily, high-resolution ambient CO exposure prediction model at the city scale.Methods:We developed a CO prediction model in Baltimore, Maryland, based on a spatiotemporal statistical algorithm with regulatory agency monitoring data and measurements from calibrated low-cost gas monitors. We also evaluated the contribution of three novel parameters to model performance: high-resolution meteorological data, satellite remote sensing data, and copollutant (, , and ) concentrations.Results:The CO model had spatial cross-validation (CV) and root-mean-square error (RMSE) of (ppm), respectively; the model had temporal CV and RMSE of 0.61 and , respectively. The predictions revealed spatially resolved CO hot spots associated with population, traffic, and other nonroad emission sources (e.g., railroads and airport), as well as sharp concentration decreases within short distances from primary roads.Discussion:The three novel parameters did not substantially improve model performance, suggesting that, on its own, our spatiotemporal modeling framework based on geographic features was reliable and robust. As low-cost air monitors become increasingly available, this approach to CO concentration modeling can be generalized to resource-restricted environments to facilitate comprehensive epidemiological research. https://doi.org/10.1289/EHP10889

Key demographics and psychological skills associated with adjustment to progressive Multiple Sclerosis early in the diagnosis.

Background/purposeBeing diagnosed with a progressive type of multiple sclerosis (MS) has been associated with worse psychological outcomes compared to relapsing-remitting type. Previous studies of adjustment to MS have primarily focused on relapsing-remitting type MS. The present study aims to examine psychological adjustment for people newly diagnosed with progressive multiple sclerosis.MethodsThis was a multicenter cross-sectional survey of 189 people newly diagnosed with progressive MS. A composite measure of psychological adjustment was created from questionnaires measuring psychological distress, positive affect, perceived-stress, life satisfaction and self-concept. Predictor variables included coping strategies, social support, relationship with partner, psychological vulnerability, MS-related beliefs, and responses to symptoms. Data were analysed using a regularised regression model to indicate which group of all variables are associated with adjustment.ResultsPeople who were older (b = 0.17(0.07), p = 0.02), in employment (b = 0.40 (0.17), p = 0.01), and with lower illness severity (b = −0.24 (0.08), p = 0.001) showed better adjustment. Based on a Lasso regression, the most important psychological and demographic variables associated with lower adjustment (out-of-sample cross-validation R2 = 62.6%) were lower MS self-efficacy and higher avoidance, cognitive vulnerability, embarrassment avoidance, conflict, helplessness, and secondary progressive MS type.Conclusions and implicationsHelping newly diagnosed people to find ways to tolerate anxiety-causing situations by encouraging acceptance may help people adjust to progressive MS by lowering their avoidance. Further, building confidence in managing the illness and addressing relationship issues are key focus areas in psychological interventions for people with progressive multiple sclerosis.

Estimating daily ground-level NO2 concentrations over China based on TROPOMI observations and machine learning approach

Nitrogen dioxide (NO2) is an important target for monitoring atmospheric quality. Deriving ground-level NO2 concentrations with much finer resolution, it requires high-resolution satellite tropospheric NO2 column as input and a reliable estimation algorithm. This paper aims to estimate the daily ground-level NO2 concentrations over China based on machine learning models and the TROPOMI NO2 data with high spatial resolution. In this study, four tree-based algorithm machine learning models, decision trees (DT), gradient boost decision tree (GBDT), random forest (RF) and extra-trees (ET), were used to estimate ground-level NO2 concentrations. In addition to considering many influencing factors of the ground-level NO2 concentrations, we especially introduced simplified temporal and spatial information into the estimation models. The results show that the extra-trees with spatial and temporal information (ST-ET) model has great performance in estimating ground-level NO2 concentrations with a cross-validation R2 of 0.81 and RMSE of 3.45 μg/m3 in test datasets. The estimated results for 2019 based on the ST-ET model achieves a satisfactory accuracy with a cross-validation R2 of 0.86 compared with the other models. Through time-space analysis and comparison, it was found that the estimated high-resolution results were consistent with the ground observed NO2 concentrations. Using data from January 2020 to test the prediction power of the models, the results indicate that the ST-ET model has a good performance in predicting ground-level NO2 concentrations. Taking four ground-level NO2 concentrations hotspots as examples, the estimated ground-level NO2 concentrations and ground-based observation data during the coronavirus disease (COVID-19) pandemic were lower compared with the same period in 2019. The findings offer a solid solution for accurately and efficiently estimating ground-level NO2 concentrations by using satellite observations, and provide useful information for improving our understanding of the regional atmospheric environment.

Ensemble averaging using remote sensing data to model spatiotemporal PM10 concentrations in sparsely monitored South Africa

There is a paucity of air quality data in sub-Saharan African countries to inform science driven air quality management and epidemiological studies. We investigated the use of available remote-sensing aerosol optical depth (AOD) data to develop spatially and temporally resolved models to predict daily particulate matter (PM10) concentrations across four provinces of South Africa (Gauteng, Mpumalanga, KwaZulu-Natal and Western Cape) for the year 2016 in a two-staged approach. In stage 1, a Random Forest (RF) model was used to impute Multiangle Implementation of Atmospheric Correction AOD data for days where it was missing. In stage 2, the machine learner algorithms RF, Gradient Boosting and Support Vector Regression were used to model the relationship between ground-monitored PM10 data, AOD and other spatial and temporal predictors. These were subsequently combined in an ensemble model to predict daily PM10 concentrations at 1 km × 1 km spatial resolution across the four provinces. An out-of-bag R2 of 0.96 was achieved for the first stage model. The stage 2 cross-validated (CV) ensemble model captured 0.84 variability in ground-monitored PM10 with a spatial CV R2 of 0.48 and temporal CV R2 of 0.80. The stage 2 model indicated an optimal performance of the daily predictions when aggregated to monthly and annual means. Our results suggest that a combination of remote sensing data, chemical transport model estimates and other spatiotemporal predictors has the potential to improve air quality exposure data in South Africa's major industrial provinces. In particular, the use of a combined ensemble approach was found to be useful for this area with limited availability of air pollution ground monitoring data.

Hologram QSAR and Topomer CoMFA Study on Naphthyridone Derivatives as ATAD2 Bromodomain Inhibitors

Background: ATAD2 is closely related to the occurrence and proliferation of many tumors. Thus, exploring ATAD2 inhibitors is greatly significant for the prevention and treatment of tumors. In this study, the quantitative structure–activity relationship (QSAR) analyses of 57 naphthyridone derivatives were conducted using hologram quantitative structure–activity relationship (HQSAR) and topomer comparative molecular field analysis (topomer CoMFA). Method: The 57 naphthyridone derivatives were divided into the training (44 derivatives) and testing (13 derivatives) sets. HQSAR and topomer CoMFA models were obtained by applying the SYBYL-X software and validated using various validation parameters. Contribution maps from the best HQSAR model and the contour maps from the best topomer CoMFA model were analyzed. Results: The most effective HQSAR model exhibited significant cross-validated (q2 = 0.872) and non cross-validated (r2 = 0.972) correlation coefficients, and the most effective topomer CoMFA model had q2 = 0.861 and r2 = 0.962. Several external validation parameters, such as , , , , and , were used to calculate the correlation coefficients of the test set samples and validate both models. The result exhibited a powerful predictive capability. Graphical results from HQSAR and topomer CoMFA were validated by the binding mode in the crystal structure. Conclusion: The models may be beneficial to enhance the understanding of the structure–activity relationships for this class of compounds and also provide useful clues for the design of potential ATAD2 bromodomain inhibitors.

Choosing Feature Selection Methods for Spatial Modeling of Soil Fertility Properties at the Field Scale

With the growing availability of environmental covariates, feature selection (FS) is becoming an essential task for applying machine learning (ML) in digital soil mapping (DSM). In this study, the effectiveness of six types of FS methods from four categories (filter, wrapper, embedded, and hybrid) were compared. These FS algorithms chose relevant covariates from an exhaustive set of 1049 environmental covariates for predicting five soil fertility properties in ten fields, in combination with ten different ML algorithms. Resulting model performance was compared by three different metrics (R2 of 10-fold cross validation (CV), robustness ratio (RR; developed in this study), and independent validation with Lin’s concordance correlation coefficient (IV-CCC)). FS improved CV, RR, and IV-CCC compared to the models built without FS for most fields and soil properties. Wrapper (BorutaShap) and embedded (Lasso-FS, Random forest-FS) methods usually led to the optimal models. The filter-based ANOVA-FS method mostly led to overfit models, especially for fields with smaller sample quantities. Decision-tree based models were usually part of the optimal combination of FS and ML. Considering RR helped identify optimal combinations of FS and ML that can improve the performance of DSM compared to models produced from full covariate stacks.

Machine Learning-Based Approach Using Open Data to Estimate PM2.5 over Europe

Air pollution is currently considered one of the most serious problems facing humans. Fine particulate matter with a diameter smaller than 2.5 micrometres (PM2.5) is a very harmful air pollutant that is linked with many diseases. In this study, we created a machine learning-based scheme to estimate PM2.5 using various open data such as satellite remote sensing, meteorological data, and land variables to increase the limited spatial coverage provided by ground-monitors. A space-time extremely randomised trees model was used to estimate PM2.5 concentrations over Europe, this model achieved good results with an out-of-sample cross-validated R2 of 0.69, RMSE of 5 μg/m3, and MAE of 3.3 μg/m3. The outcome of this study is a daily full coverage PM2.5 dataset with 1 km spatial resolution for the three-year period of 2018–2020. We found that air quality improved throughout the study period over all countries in Europe. In addition, we compared PM2.5 levels during the COVID-19 lockdown during the months March–June with the average of the previous 4 months and the following 4 months. We found that this lockdown had a positive effect on air quality in most parts of the study area except for the United Kingdom, Ireland, north of France, and south of Italy. This is the first study that depends only on open data and covers the whole of Europe with high spatial and temporal resolutions. The reconstructed dataset will be published under free and open license and can be used in future air quality studies.

Contributors to Serum NfL Levels in People without Neurologic Disease.

To assess the effects of demographics, lifestyle factors, and comorbidities on serum neurofilament light chain (sNfL) levels in people without neurologic disease and establish demographic-specific reference ranges of sNfL. The National Health and Nutrition Examination Survey (NHANES) is a representative sample of the US population in which detailed information on demographic, lifestyle, routine laboratory tests, and overall health status are systematically collected. From stored serum samples, we measured sNfL levels using a novel high-throughput immunoassay (Siemens Healthineers). We evaluated the predictive capacity of 52 demographic, lifestyle, comorbidity, anthropometric, or laboratory characteristics in explaining variability in sNfL levels. Predictive performance was assessed using cross-validated R2 (R2 cv ) and forward selection was used to obtain a set of best predictors of sNfL levels. Adjusted reference ranges were derived incorporating characteristics using generalized additive models for location, scale, and shape. We included 1,706 NHANES participants (average age: 43.6 ± 14.8 y; 50.6% male, 35% non-white) without neurological disorders. In univariate models, age explained the most variability in sNfL (R2 cv = 26.8%). Multivariable prediction models for sNfL contained three covariates (in order of their selection): age, creatinine, and glycosylated hemoglobin (HbA1c) (standardized β-age: 0.46, 95% confidence interval [CI]: 0.43, 0.50; creatinine: 0.18, 95% CI: 0.13, 0.22; HbA1c: 0.09, 95% CI: 0.06, 0.11). Adjusted centile curves were derived incorporating identified predictors. We provide an interactive R Shiny application to translate our findings and allow other investigators to use the derived centile curves. Results will help to guide interpretation of sNfL levels as they relate to neurologic conditions. ANN NEUROL 2022;92:688-698.

Hourly Seamless Surface O3 Estimates by Integrating the Chemical Transport and Machine Learning Models in the Beijing-Tianjin-Hebei Region

Surface ozone (O3) is an important atmospheric trace gas, posing an enormous threat to ecological security and human health. Currently, the core objective of air pollution control in China is to realize the joint treatment of fine particulate matter (PM2.5) and O3. However, high-accuracy near-surface O3 maps remain lacking. Therefore, we established a new model to determine the full-coverage hourly O3 concentration with the WRF-Chem and random forest (RF) models combined with anthropogenic emission data and meteorological datasets. Based on this method, choosing the Beijing-Tianjin-Hebei (BTH) region in 2018 as an example, full-coverage hourly O3 maps were generated at a horizontal resolution of 9 km. The performance evaluation results indicated that the new model is reliable with a sample (station)-based 10-fold cross-validation (10-CV) R2 value of 0.94 (0.90) and root mean square error (RMSE) of 14.58 (19.18) µg m−3. In addition, the estimated O3 concentration is accurately determined at varying temporal scales with sample-based 10-CV R2 values of 0.96, 0.98 and 0.98 at the daily, monthly, and seasonal scales, respectively, which is highly superior to traditional derivation algorithms and other techniques in previous studies. An initial increase and subsequent decrease, which constitute the diurnal variation in the O3 concentration associated with temperature and solar radiation variations, were captured. The highest concentration reached approximately 112.73 ± 9.65 μg m−3 at 15:00 local time (1500 LT) in the BTH region. Summertime O3 posed a high pollution risk across the whole BTH region, especially in southern cities, and the pollution duration accounted for more than 50% of the summer season. Additionally, 43 and two days exhibited light and moderate O3 pollution, respectively, across the BTH region in 2018. Overall, the new method can be beneficial for near-surface O3 estimation with a high spatiotemporal resolution, which can be valuable for research in related fields.

A Multiobjective Optimization Approach to Pulmonary Rehabilitation Effectiveness in COPD

Chronic obstructive pulmonary disease (COPD) is a common disease that accounts for a significant individual and societal burden. Pulmonary rehabilitation (PR) is a key management strategy but it is highly inaccessible, making prioritisation highly needed. This study aimed to determine and optimize predictive models of PR outcomes and build a tool to help healthcare professionals in their clinical decision-making about PR prioritisation. Data from patients who performed a 12-week community-based PR programme were analysed. Exercise capacity with the six-minutes walk test distance (6MWD), isometric quadriceps muscle strength with the handheld dynamometry (QMS) and dyspnoea with the modified Medical Research Council dyspnoea scale (mMRC) were assessed before and after PR. Multiple linear regression models were determined based on the Akaike information criteria and a cross-validation method. The resultant multiobjective problem was solved using the Nondominated Sorting Genetic Algorithm-II. R Shiny package was used to create a web-based user interface. Data from 95 patients with COPD (median age of 69 years, 19 female and generally overweight), resulted in linear predictive models for the post-pre difference of the 6MWD, QMS and mMRC with cross-validation R2 of 0.49, 0.53 and 0.51, respectively. 6MWD and mMRC were common statistically significant predictors. Pareto front patients were obese ex-smoker women that do not do long-term oxygen therapy and that performed PR. The distance to the Pareto front along with the estimates given by our models are easily obtained using the designed R Shiny interface and may help healthcare professionals decide on the prioritisation to PR programmes.