Welcome to the Optibrium Community

Forgot login?




Popular Downloads

Flavour and Fragrance Models: Skin Absorption Modelling

Thursday, 26 April 2018 15:18

Fragrance materials are widely used in cosmetics and other consumer products. The safety assessment of these ingredients includes skin absorption as this is an important parameter in estimating systemic exposure. In the current safety assessment process a compound is assumed as having 100% skin absorption when experimental data are lacking. Hence better estimates for absorption will help provide a better assessment of new fragrances and topically applied chemicals.


A model has been created with reference to published work from Shen et al directed at predicting Jmax (a theoretically achieved dose based on Fick’s first law of diffusion) where Jmax = Kp * Cwater on a data set of fragrance-like molecules. The 131 Jmax values are skewed but LogJmax are more normally distributed and so LogJmax modelled in Auto-Modeller using an 80:10:10% split of the data.

The values of LogCwater and LogK(oct/water) from the paper can be modelled successfully with the LogS and LogP models (available within StarDrop and ADME QSAR module). The Kp value in the paper is derived from a consensus of 7 models for LogK(oct/water), this property can be calculated directly. The Shen paper contains experimental and estimated values for Kp and was combined with Kp data derived from Alves et al to give a data set of 202 compounds. This set was split in 70:15:15% to generate a model for LogKp.


The best performing model for LogJmax was generated using the Gaussian Process methodology with rescaled forward variable selection (GP-RFVS).

LogJmax model

Modelling method Training set Validation set Test set
GP-RFVS 0.86 0.45 0.81 0.53 0.85 0.45

The best performing model for LogKp was generated using a GA-RBF model

LogKp Model

Modelling method Training set Validation set Test set
GA-RBF 1 0.005 0.80 0.51 0.71 0.52


Flavour and Fragrance Models: Leffingwell Odour Threshold

Thursday, 26 April 2018 14:25

To be smelt a compound has to have a low enough vapour pressure to be in the gaseous state and have sufficient affinity/efficacy at olfactory receptors. A related measure of volatility is odour threshold.


A data set was collected from the web pages (accessed Jan2018) created by Dr John C. Leffingwell. It details the odour thresholds of various chemical classes where chirality modifies the odour descriptions of the enantiomeric pairs.

The full data set contains over 700 enantiomeric pairs, however, only 422 compounds had suitable data for modelling the odour threshold quantified in ppb. The data range for this set was extremely large, covering more than 10 orders of magnitude and so the log of the ppb data was modelled. Where compounds were listed as “odourless” a value of 7 was used as a default; ppb values of ‘-3’ and ‘1’ were used to classify the set. As the StarDrop descriptors are unaware of chirality the category boundaries were chosen to cope with the changes in threshold which were solely due to changes at the chiral centre(s).


The best model produced was a Random Forest model.

Modelling method Training set Test set
Kappa Accuracy Kappa Accuracy
RF Classification 0.79 0.86 0.58 0.73



Monday, 19 March 2018 04:08

This script was developed by Travis Hesketh as part of an undergraduate industrial placement. It explores combinations of assay data for the same target within ChHEMBL to identify those assays that are most likely to result in a high quality training set of consistent data.


Latest Forums

Read more >