HSP Examples: In-silico testing

It is a reasonable idea that you could screen chemicals for their tendency to cause, say, eye irritation, based on their solubility profile. The chemical cannot cause irritation if it cannot dissolve into the relevant parts of the eye. It is, therefore, a reasonable hypothesis that the HSP of a chemical will have a correlation to the Draize test for eye irritation. With one exception, known in advance, that strong acids (pH < 2) and strong bases (pH > 11) damage the eye independently of their solubility profile.

A team led by Martin Andersson at RISE in Sweden1 [available at this Open Access link] tested this idea via a careful program based on HSPiP, plus an external tool for predicting pKa values.There is a large Draize test dataset of chemicals with their SMILES structure. The team first ran the SMILES through HSPiP's File Check function to ensure that chemicals were in a meaningful "applicability domain" eliminating those chemicals that were outside reasonable HSP prediction e.g. salts, unusual atoms, very large molecules. The File Check function was added to HSPiP as a request from the Andersson team who actively tested and helped debug the code. Then predicted strong acid/bases were removed. Finally, using a genetic algorithm, the team found a strong predictive correlation between HSP and Draize values, using standard statistical methods to check for fit quality and predictability, checking both for true & false positives and true & false negatives.

The overall quality of the prediction is good, allowing a rapid in-silico check of any (reasonable) pure liquid directly from its SMILES.

The HSP Sphere

The HSP sphere of Draize chemicalsInterestingly, the data allowed the team to fit an HSP sphere to the Draize data, with the centre of activity being [18.4, 10.2, 11.0] with a radius of 8.6. They used a sophisticated sphere optimization technique to obtain these values.

To create an image of the sphere for this page, Martin Andersson kindly put the data into HSPiP and created an acceptable (and similar) fit using the standard HSPiP methodology.

1 Martin Andersson, Ulf Norinder, Swapnil Chavan and Ian Cotgreave, In Silico Prediction of Eye Irritation Using Hansen Solubility Parameters and Predicted pKa Values, DOI: 10.1177/02611929231175676

The official site of Hansen Solubility Parameters and HSPiP software.