4.0 Conclusions

In this paper we have reported the use of a quantum chemical-based approach to assess the skin sensitization potential of chemicals from the Schiff base domain. We have evaluated the mechanistic profile associated with 22 SB substrates using a model consisting of two methylamine and two water molecules. We find that calculating the full reaction profile for the chemicals is important as the substrates can often react in more than one position, while also allowing for mechanistic exceptions to be uncovered. We find that the use of a single computed descriptor, namely the rate determining barrier to formation of the SB product can help us to separate sensitizer and non-sensitizer. A RDS barrier of ~28 kcal/mol indicate that a molecule is unlikely to act as a sensitizer. We also observed that compounds with low barriers, but higher logP values show reduced sensitization prompting us to generate a 2 parameter QMM.
A QMM equation established suggests that SB of lower logP have a greater propensity to react resulting in r2 of 0.50-0.60. The predicted RDS and logP establish SAR guidelines to rationalize the skin sensitization potential. The RDS barriers for aldehydes, ketone, 1,2 and 1,3 diones broadly decrease in that order, in line with their increasing experimental sensitivity. These findings agree with experimental based observations in the literature and point to the value computational methods can play in skin sensitization predictions. We find that the rate determining barrier and the computed lipophilicity can be used to estimate the skin-sensitization of unknown compounds. This orthogonal source of information could prove useful in consensus based predictions of likely sensitization potential.[21, 26]
The results presented here show that 3D quantum chemical simulation of SB chemicals, while useful, will lead to the mischaracterization of some compounds. This is not so different to the variation observed between the different types of in vivo , in vitro and in silico methods reported to date which show predictions accuracies of no more than 70-80%. This is simply a reflection of the complex event being simulated, a multitude of potential protein targets, and the fact that the molecules may function in the form of a metabolite rather than the dosed substrate. The utility of such simulations is that physical insight and understanding can be garnered which could prove useful, especially when combined in the so–called weight of evidence approach with other methods.