OWSum: algorithmic odor prediction and insight into structure-odor relationships

Schicker, Doris; Singh, Satnam; Freiherr, Jessica; Grasskamp, Andreas T.

doi:10.1186/s13321-023-00722-y

“…Lötsch et al [25] showed applications of olfactometric data with a set of unsupervised and supervised algorithms for pattern-based odor detection and recognition, odor prediction from physicochemical properties of volatile molecules, and knowledge discovery in publicly available big databases. Two different approaches to predict the structure-odor relationship with machine learning were conducted by Schicker et al [26]-who developed a classification algorithm that quantitatively assigns structural patterns to odors-and Bo et al [27], who used deep learning on the structural features for a binary two-class prediction of the odors. It was possible for Lee et al [28] to generate a principal odor map by constructing a message passing an artificial neural network (NN) to map chemical structures to odor percepts that enable odor quality prediction with human-level odor description performance and outperform chemoinformatic models.…”

Section: Modeling Aroma Partitioningmentioning

confidence: 99%

Using a Machine Learning Regression Approach to Predict the Aroma Partitioning in Dairy Matrices

Anker,

Borsum,

Zhang

et al. 2024

Processes

3

0

View full text Add to dashboard Cite

Aroma partitioning in food is a challenging area of research due to the contribution of several physical and chemical factors that affect the binding and release of aroma in food matrices. The partition coefficient measured by the Kmg value refers to the partition coefficient that describes how aroma compounds distribute themselves between matrices and a gas phase, such as between different components of a food matrix and air. This study introduces a regression approach to predict the Kmg value of aroma compounds of a wide range of physicochemical properties in dairy matrices representing products of different compositions and/or processing. The approach consists of data cleaning, grouping based on the temperature of Kmg analysis, pre-processing (log transformation and normalization), and, finally, the development and evaluation of prediction models with regression methods. We compared regression analysis with linear regression (LR) to five machine-learning-based regression algorithms: Random Forest Regressor (RFR), Gradient Boosting Regression (GBR), Extreme Gradient Boosting (XGBoost, XGB), Support Vector Regression (SVR), and Artificial Neural Network Regression (NNR). Explainable AI (XAI) was used to calculate feature importance and therefore identify the features that mainly contribute to the prediction. The top three features that were identified are log P, specific gravity, and molecular weight. For the prediction of the Kmg in dairy matrices, R2 scores of up to 0.99 were reached. For 37.0 °C, which resembles the temperature of the mouth, RFR delivered the best results, and, at lower temperatures of 7.0 °C, typical for a household fridge, XGB performed best. The results from the models work as a proof of concept and show the applicability of a data-driven approach with machine learning to predict the Kmg value of aroma compounds in different dairy matrices.

show abstract

“…Such representations have therefore been used extensively with machine learning in previous works, like those to generate new molecular structures [13][14][15][16][17]. Additionally, other encoding schemes, such as molecular graphs [15,[18][19][20], have been widely combined with machine learning to predict molecular properties, like toxicity [21,22], medical activity in drug discovery [23][24][25][26], or even to predict the odor of molecules [27][28][29][30][31].…”

Section: Introductionmentioning

confidence: 99%

Classification of substances by health hazard using deep neural networks and molecular electron densities

Singh,

Zeh,

Freiherr

et al. 2024

J Cheminform

Self Cite

0

View full text Add to dashboard Cite

In this paper we present a method that allows leveraging 3D electron density information to train a deep neural network pipeline to segment regions of high, medium and low electronegativity and classify substances as health hazardous or non-hazardous. We show that this can be used for use-cases such as cosmetics and food products. For this purpose, we first generate 3D electron density cubes using semiempirical molecular calculations for a custom European Chemicals Agency (ECHA) subset consisting of substances labelled as hazardous and non-hazardous for cosmetic usage. Together with their 3-class electronegativity maps we train a modified 3D-UNet with electron density cubes to segment reactive sites in molecules and classify substances with an accuracy of 78.1%. We perform the same process on a custom food dataset (CompFood) consisting of hazardous and non-hazardous substances compiled from European Food Safety Authority (EFSA) OpenFoodTox, Food and Drug Administration (FDA) Generally Recognized as Safe (GRAS) and FooDB datasets to achieve a classification accuracy of 64.1%. Our results show that 3D electron densities and particularly masked electron densities, calculated by taking a product of original electron densities and regions of high and low electronegativity can be used to classify molecules for different use-cases and thus serve not only to guide safe-by-design product development but also aid in regulatory decisions. Scientific contribution We aim to contribute to the diverse 3D molecular representations used for training machine learning algorithms by showing that a deep learning network can be trained on 3D electron density representation of molecules. This approach has previously not been used to train machine learning models and it allows utilization of the true spatial domain of the molecule for prediction of properties such as their suitability for usage in cosmetics and food products and in future, to other molecular properties. The data and code used for training is accessible at https://github.com/s-singh-ivv/eDen-Substances.

show abstract

“…Such representations have therefore been used extensively with machine learning in previous works, like those to generate new molecular structures [13][14][15][16][17]. Additionally, other encoding schemes, such as molecular graphs [15,[18][19][20], have been widely combined with machine learning to predict molecular properties, like toxicity [21,22], medical activity in drug discovery [23][24][25][26], or even to predict the odor of molecules [27][28][29][30][31], among others actions.…”

Section: Introductionmentioning

confidence: 99%

Classification of substances by health hazard using deep neural networks and molecular electron densities

Singh,

Zeh,

Freiherr

et al. 2023

Preprint

Self Cite

1

0

View full text Add to dashboard Cite

In this paper we present a method that allows leveraging 3D electron density information to train a deep neural network pipeline to segment regions of high, medium and low electronegativity and classify substances as health hazardous or non-hazardous. We show that this can be used for use-cases such as cosmetics and food products. For this purpose, we first generate 3D electron density cubes using semiempirical molecular calculations for a custom European Chemical Agency (ECHA) subset consisting of substances labelled as hazardous and non-hazardous for cosmetic usage. Together with their 3-class electronegativity maps we train a modified 3D-UNet with electron density cubes to segment reactive sites in molecules and classify substances with an accuracy of 78.1%. We perform the same process on a custom food dataset (CompFood) consisting of hazardous and non-hazardous substances compiled from European Food Safety Authority (EFSA) OpenFoodTox, Food and Drug Administration (FDA) Generally Recognized as Safe (GRAS) and FooDB datasets to achieve a classification accuracy of 64.1%. Our results show that 3D electron densities and particularly masked electron densities denoting regions of high and low reactivity can be used to classify molecules for different use-cases and thus serve not only to guide safe-by-design product development but also aid in regulatory decisions.

show abstract

OWSum: algorithmic odor prediction and insight into structure-odor relationships

Cited by 10 publications

References 55 publications

Using a Machine Learning Regression Approach to Predict the Aroma Partitioning in Dairy Matrices

Using a Machine Learning Regression Approach to Predict the Aroma Partitioning in Dairy Matrices

Classification of substances by health hazard using deep neural networks and molecular electron densities

Classification of substances by health hazard using deep neural networks and molecular electron densities

Contact Info

Product

Resources

About