Figure 2. Plot of the SNAr reaction profiles for compounds 2 , 4 and 18 .
It should be acknowledged that such analyzes are further complication by the fact that the LLNA response is not always a result of direct sensitization the molecule itself, rather an effect of reactive metabolites. This issue further complicates the construction and validation of computational methodologies such as these reported here.[21, 64] In particular, it is known that 1,2 diones such as compound 16 , glyoxal are not commonly encountered because it can form a hydrate, which can then undergo oligomerization.

3.2 Reaction Barriers vs LLNA pEC3

The goal of the present work is to determine whether it is possible to attribute the skin sensitization to the predicted chemical reactivity of the functional groups present the molecules1-22 . We first assessed whether a relationship existed between the predicted barriers to reaction and the skin sensitivity. The expectation was that TS2 should show the most significant correlation with the LLNA pEC3 given that this was almost exclusively the rate determining barrier for the molecules studied here. Indeed we did see the strongest correlation between the TS2 (lowest value of the reported positions if more than one) and the pEC3. However, when all 22 compounds were included, the observed correlation coefficient squared (r2) was just 0.16. This prompted us to re-evaluate the data by excluding compound 18 , which is a predicted to be a strong sensitizer via the SNAr domain, and the 1,2 dione compound (5, 9, 16, 17, 20 ), which potentially exist in their hydrated form. This is similar to the approach of Roberts et al. who focused on compounds with a single reactive group consisting primarily of aldehydes and ketones and may point to a need for a more restricted domain of applicability.[32]
Re-analysis of the correlation between the 16 remaining compounds and the pEC3 reveals an improved, albeit weak correlation, with an r2 of 0.36. It is found as the predicted barrier to reaction decreases, the experimental sensitization is found to increase in line with expectation (Figure 2). It can also be seen from Figure 3 there is a lipophilicity effect operating within the dataset. Compounds, which have been colored by logP, show a noticeable trend in that, those with a higher logP generally have lower pEC3 for a given barrier. For example, the 1,3 dione 8 has a predicted barrier of approximately 22 kcal/mol compared to 19 which has a barrier of ~25 kcal/mol (19 a). This would suggest that 8 should be a stronger sensitizer than 19 , whereas in fact the opposite is observed (pEC3 of 0.84 vs 1.97, respectively). The predicted logP of 8 is 3.93 while that of 19 is 1.28 suggesting that a lower barrier combined with a low logP leads to increased sensitization for this set of diverse compounds.
This observation apparently contradicts the findings of the earlier QSAR study of Roberts et al. on Schiff bases.[32] However, since this publication, the authors have reported an extensive analysis of 525 substances and note that logP is not simply a surrogate for permeability and that higher logP values do not result in higher sensitization on a diverse set of compounds.[65] The result is also consistent with observations that the Schiff base, formaldehyde (logP=-0.51), which is known to be highly sensitizing in sensitization based-assays. [66] It should be noted that we are using a somewhat larger (22 vs 16 compounds), and more diverse set of chemicals (70% were aldehydes in the latter study). Indeed evidence for similar series specific-logP behavior can be seen from the non-linear relationship between logP and pEC3 for azlactones of the acyl domain.[5]