Visual Analysis of Odor Interaction Based on Support Vector Regression Method

Utilisation of QSPR ODT modelling and odour vector modelling to predict Cannabis sativa odour

Cannabis flower odour is an important aspect of product quality as it impacts the sensory experience when administered, which can affect therapeutic outcomes in paediatric patient populations who may reject unpalatable products. However, the cannabis industry has a reputation for having products with inconsistent odour descriptions and misattributed strain names due to the costly and laborious nature of sensory testing. Herein, we evaluate the potential of using odour vector modelling for predicting the odour intensity of cannabis products. Odour vector modelling is proposed as a process for transforming routinely produced volatile profiles into odour intensity (OI) profiles which are hypothesised to be more informative to the overall product odour (sensory descriptor; SD). However, the calculation of OI requires compound odour detection thresholds (ODT), which are not available for many of the compounds present in natural volatile profiles. Accordingly, to apply the odour vector modelling process to cannabis, a QSPR statistical model was first produced to predict ODT from physicochemical properties. The model presented herein was produced by polynomial regression with 10-fold cross-validation from 1,274 median ODT values to produce a model with R2 = 0.6892 and a 10-fold R2 = 0.6484. This model was then applied to terpenes which lacked experimentally determined ODT values to facilitate vector modelling of cannabis OI profiles. Logistic regression and k-means unsupervised cluster analysis was applied to both the raw terpene data and the transformed OI profiles to predict the SD of 265 cannabis samples and the accuracy of the predictions across the two datasets was compared. Out of the 13 SD categories modelled, OI profiles performed equally well or better than the volatile profiles for 11 of the SD, and across all SD the OI data was on average 21.9% more accurate (p = 0.031). The work herein is the first example of the application of odour vector modelling to complex volatile profiles of natural products and demonstrates the utility of OI profiles for the prediction of cannabis odour. These findings advance both the understanding of the odour modelling process which has previously only been applied to simple mixtures, and the cannabis industry which can utilise this process for more accurate prediction of cannabis odour and thereby reduce unpleasant patient experiences.

Evaluation of Odor Prediction Model Performance and Variable Importance according to Various Missing Imputation Methods

The aim of this study is to ascertain the most suitable model for predicting complex odors using odor substance data that has a small number of data and a large number of missing data. First, we compared the data removal and imputation methods, and the method of imputing missing data was found to be more effective. Then, in order to recommend a suitable model, we created a total of 126 models (missing imputation: single imputation, multiple imputations, K-nearest neighbor imputation; data preprocessing: standardization, principal component analysis, partial least square; and predictive method: multiple regression, machine learning, deep learning) and compared them using R2 and mean absolute error (MAE) values. Finally, we investigated variable importance using the best prediction model. The results identified the best model as a combination of multivariate imputation using Bayesian ridge as the missing imputation method, standardization for data preprocessing, and an extremely randomized tree as the predictive method. Among the odor compounds, Methyl mercaptan, acetic acid, and dimethyl sulfide were identified as the most important odor compounds in predicting complex odors.

Application of Machine Learning for Fenceline Monitoring of Odor Classes and Concentrations at a Wastewater Treatment Plant

The development of low-cost sensors, the introduction of technical performance specifications, and increasingly effective machine learning algorithms for managing big data have led to a growing interest in the use of instrumental odor monitoring systems (IOMS) for odor measurements from industrial plants. The classification and quantification of odor concentration are the main goals of IOMS installed inside industrial plants in order to identify the most important odor sources and to assess whether the regulatory thresholds have been exceeded. This paper illustrates the use of two machine learning algorithms applied to the concurrent classification and quantification of odors. Random Forest was employed, which is a machine learning algorithm that thus far has not been used in the field of odor quantification and classification for complex industrial situations. Furthermore, the results were compared with commonly used algorithms in this field, such as artificial neural network (ANN), which was here employed in the form of a deep neural network. Both techniques were applied to the data collected from an IOMS installed for fenceline monitoring at a wastewater treatment plant. Cohen’s kappa and Normalized RMSE are used as specifical performance indicators for classification and regression: the indicators were calculated for the test dataset, and the results were compared with data in the literature obtained in contexts of similar complexity. A Cohen’s kappa of 97% was reached for the classification task, while the best Normalized RMSE, namely 4%, for the interval 20–2435 ouE/m³ was obtained with Random Forest.