Flavour and fragrance models: Leffingwell odour threshold
To be smelt a compound has to have a low enough vapour pressure to be in the gaseous state and…
Auto-Modeller doesn’t currently detect outliers when building models, it is dependent on a user’s own judgement.
The Gaussian Processes methods do employ a sophisticated process of detecting conflicting values and they will put a higher uncertainty onto this information. So, these methods are more robust when it comes to handling outliers. However, this detection isn’t reported in the Session Details.
For any method, the greater the number of compounds that conflict with the outlier, the less the outlier will influence the model. So, the robustness of the model does depend on the size of the data set and the proportion of outliers. Having a large and reliable set of data isn’t always possible, but this is the reason why we have the independent validation and test sets.
It is also worth noting that the uncertainty you’ve defined in your data isn’t taken into consideration when building a model.
More models resources
HTS promiscuity alerts model
Screen for HTS promiscuity with StarDrop HTS Promiscuity Alerts.
PAINS Model
Baell and Holloway published a set of substructure filters for removal of what they termed “Pan Assay Interference Compounds (PAINS)” from…