Functional recognition of structure-diverse odor molecules in drinking water based on QSOR study. (November 2018)

• Record Type: Journal Article
• Authors: Yu, Jianwei; Zhang, Li; Zhang, Ying; An, Wei; Guo, Qingyuan; Zhao, Yu; Yang, Min
• Is Part Of: Chemosphere. Volume 211(2018)
• Page Start: 371
• Page End: 378
• Publication Date: 2018-11
• Subjects: Odorant structure -- Odor threshold -- Partial least squares -- Binary QSOR; Pollution -- Periodicals; Pollution -- Physiological effect -- Periodicals; Environmental sciences -- Periodicals; Atmospheric chemistry -- Periodicals; 551.511
• DOI: 10.1016/j.chemosphere.2018.07.149
• Languages: English
• ISSNs: 0045-6535

Abstract

Abstract: Taste and odor problems in drinking water have long been plaguing many water utilities and the public. Even though many odorants have been reported, up to now, identification of the odor-causing compounds is still a challenge for the water industry. In this study, 22 typical reported odor compounds with similar odor characteristics were selected as the training set to build the linear quantitative structure odor relationship (QSOR) model by the partial least squares (PLS) method. The logarithm of the odor threshold (OT) value divided by the molecular weight of the responsible compound ( p OT) was selected as the response descriptor to express odor characteristics. The resulting good statistical results, with R 2 (correlation coefficient) = 0.8988, RMSE (root mean square error) = 0.4374, XR 2 (cross-validated correlation coefficient) = 0.8133, and XRMSE (cross-validated root mean square error) = 0.5993, indicate that the odor thresholds of potential odorants with similar or distinguishable odors could be predicted using the model with corresponding descriptor data of known-structure odorants. Moreover, external validation was also conducted using the nonlinear binary QSOR method, where the overall binary QSOR accuracy remained stable (around 90%) regardless of the chosen threshold values. By using the validated QSOR model, the p OT of the set of 8 test compounds was successfully predicted with good correlation to their experimental p OT values. This study could provide a novel and convenient way to screen the potential odorants from innumerable candidate chemicals. Highlights: A novel and convenient way to screen the potential odorants was provided. One linear QSOR model with high predictive quality was built. Nonlinear binary QSOR method was applied for the validation of the predictive ability.