Development of an embedded molecular structure-based model for prediction of micropollutant treatability in a drinking water treatment plant by machine learning from three years monitoring data.

In this study, an autoencoder-based molecular structure embedding model was developed to predict treatability of micropollutant in a drinking water treatment plant (DWTP) by machine learning using 69 micropollutants monitoring data at 18 DWTPs for three years. The molecular structure, which contains physicochemical characteristics, was embedded as a fixed-length vector that is advantageous for data-driven analysis and machine learning. First, the molecular structure of the micropollutants was converted to a sequence of tokens using the simplified molecular-input line-entry system (SMILES) pair encoding tokenizer, a frequency-based tokenization method. It was then compressed into fixed-length vectors using an autoencoder trained on various molecular structures within the Chemical Entities of Biological Interest. To validate the proposed models, a binary classification of micropollutant treatability was performed using the embedded molecular structure of micropollutants with various external features, such as concentration, season, and the presence of specific drinking water treatment processes by machine learning. The accuracy of the developed model for the 69 micropollutants in this study was 0.86, and the molecular structure was determined to be the most important feature. Furthermore, an accuracy of 0.71 was obtained in external validation for pharmaceuticals and personal care products that were not used for training. This shows that the proposed embedding vector can be generalized to unseen molecules during the training process, which means that it reflects the characteristics of the molecular structures.

Exploiting temporal and nonstationary features in breathing sound analysis for multiple obstructive sleep apnea severity classification.

BACKGROUND: Polysomnography (PSG) is the gold standard test for obstructive sleep apnea (OSA), but it incurs high costs, requires inconvenient measurements, and is limited by a one-night test. Thus, a repetitive OSA screening test using affordable data would be effective both for patients interested in their own OSA risk and in-hospital PSG. The purpose of this research was to develop a four-OSA severity classification model using a patient's breathing sounds. METHODS: Breathing sounds were recorded from 83 subjects during a PSG test. There was no exclusive experimental protocol or additional recording instruments use throughout the sound recording procedure. Based on the Apnea-Hypopnea Index (AHI), which indicates the severity of sleep apnea, the subjects' sound data were divided into four-OSA severity classes. From the individual sound data, we proposed two novel methods which were not attempted in previous OSA severity classification studies. First, the total transition probability of approximated sound energy in time series, and second, the statistical properties derived from the dimension-reduced cyclic spectral density. In addition, feature selection was conducted to achieve better results with a more relevant subset of features. Then, the classification model was trained using support vector machines and evaluated using leave-one-out cross-validation. RESULTS: The overall results show that our classification model is better than existing multiple OSA severity classification method using breathing sounds. The proposed method demonstrated 79.52% accuracy for the four-class classification task. Additionally, it demonstrated 98.0% sensitivity, 75.0% specificity, and 92.78% accuracy for OSA subject detection classification with AHI threshold 5. CONCLUSIONS: The results show that our proposed method can be used as part of an OSA screening test, which can provide the subject with detailed OSA severity results from only breathing sounds.