Prediction of the Response of a Photoionization Detector to a Complex Gaseous Mixture of Volatile Organic Compounds Produced by α-Pinene Oxidation.

Photoionization detectors (PIDs) are lightweight and respond in real time to the concentrations of volatile organic compounds (VOCs), making them suitable for environmental measurements on many platforms. However, the nonselective sensing mechanism of PIDs challenges data interpretation, particularly when exposed to the complex VOC mixtures prevalent in the Earth's atmosphere. Herein, two approaches to this challenge are investigated. In the first, quantum-chemistry calculations are used to estimate photoionization cross sections and ionization potentials of individual species. In the second, machine learning models are trained on these calculated values, as well as empirical PID response factors, and then used for prediction. For both approaches, the resulting information for individual species is used to model the overall PID response to a complex VOC mixture. In complement, laboratory experiments in the Harvard Environmental Chamber are carried out to measure the PID response to the complex molecular mixture produced by α-pinene oxidation under various conditions. The observations show that the measured PID response is 15% to 30% smaller than the PID response modeled by quantum-chemistry calculations of the photoionization cross section for the photo-oxidation experiments and 15% to 20% for the ozonolysis experiments. By comparison, the measured PID response is captured within a 95% confidence interval by the use of machine learning to model the PID response based on the empirical response factor in all experiments. Taken together, the results of this study demonstrate the application of machine learning to augment the performance of a nonselective chemical sensor. The approach can be generalized to other reactive species, oxidants, and reaction mechanisms, thus enhancing the utility and interpretability of PID measurements for studying atmospheric VOCs.

Machine Learning for Ionization Potentials and Photoionization Cross Sections of Volatile Organic Compounds.

Molecular ionization potentials (IP) and photoionization cross sections (σ) can affect the sensitivity of photoionization detectors (PIDs) and other sensors for gaseous species. This study employs several methods of machine learning (ML) to predict IP and σ values at 10.6 eV (117 nm) for a dataset of 1251 gaseous organic species. The explicitness of the treatment of the species electronic structure progressively increases among the methods. The study compares the ML predictions of the IP and σ values to those obtained by quantum chemical calculations. The ML predictions are comparable in performance to those of the quantum calculations when evaluated against measurements. Pretraining further reduces the mean absolute errors (ε) compared to the measurements. The graph-based attentive fingerprint model was most accurate, for which εIP = 0.23 ± 0.01 eV and εσ = 2.8 ± 0.2 Mb compared to measurements and computed cross sections, respectively. The ML predictions for IP correlate well with both the measured IPs (R 2 = 0.88) and with IPs computed at the level of M06-2X/aug-cc-pVTZ (R 2 = 0.82). The ML predictions for σ correlated reasonably well with computed cross sections (R 2 = 0.66). The developed ML methods for IP and σ values, representing the properties of a generalizable set of volatile organic compounds (VOCs) relevant to industrial applications and atmospheric chemistry, can be used to quantitatively describe the species-dependent sensitivity of chemical sensors that use ionizing radiation as part of the sensing mechanism, such as photoionization detectors.

River Winds and Transport of Forest Volatiles in the Amazonian Riparian Ecoregion.

Volatile organic compounds (VOCs) emitted from forests are important chemical components that affect ecosystem functioning, atmospheric chemistry, and regional climate. Temperature differences between a forest and an adjacent river can induce winds that influence VOC fate and transport. Quantitative observations and scientific understanding, however, remain lacking. Herein, daytime VOC datasets were collected from the surface up to 500 m over the "Rio Negro" river in Amazonia. During time periods of river winds, isoprene, α-pinene, and ß-pinene concentrations increased by 50, 60, and 80% over the river, respectively. The concentrations at 500 m were up to 80% greater compared to those at 100 m because of the transport path of river winds. By comparison, the concentration of methacrolein, a VOC oxidation product, did not depend on river winds or height. The differing observations for primary emissions and oxidation products can be explained by the coupling of timescales among emission, reaction, and transport. This behavior was captured in large-eddy simulations with a coupled chemistry model. The observed and simulated roles of river winds in VOC fate and transport highlight the need for improved representation of these processes in regional models of air quality and chemistry-climate coupling.

Machine learning and deep learning to predict mortality in patients with spontaneous coronary artery dissection.

Machine learning (ML) and deep learning (DL) can successfully predict high prevalence events in very large databases (big data), but the value of this methodology for risk prediction in smaller cohorts with uncommon diseases and infrequent events is uncertain. The clinical course of spontaneous coronary artery dissection (SCAD) is variable, and no reliable methods are available to predict mortality. Based on the hypothesis that machine learning (ML) and deep learning (DL) techniques could enhance the identification of patients at risk, we applied a deep neural network to information available in electronic health records (EHR) to predict in-hospital mortality in patients with SCAD. We extracted patient data from the EHR of an extensive urban health system and applied several ML and DL models using candidate clinical variables potentially associated with mortality. We partitioned the data into training and evaluation sets with cross-validation. We estimated model performance based on the area under the receiver-operator characteristics curve (AUC) and balanced accuracy. As sensitivity analyses, we examined results limited to cases with complete clinical information available. We identified 375 SCAD patients of which mortality during the index hospitalization was 11.5%. The best-performing DL algorithm identified in-hospital mortality with AUC 0.98 (95% CI 0.97-0.99), compared to other ML models (P < 0.0001). For prediction of mortality using ML models in patients with SCAD, the AUC ranged from 0.50 with the random forest method (95% CI 0.41-0.58) to 0.95 with the AdaBoost model (95% CI 0.93-0.96), with intermediate performance using logistic regression, decision tree, support vector machine, K-nearest neighbors, and extreme gradient boosting methods. A deep neural network model was associated with higher predictive accuracy and discriminative power than logistic regression or ML models for identification of patients with ACS due to SCAD prone to early mortality.