Skin sensitization DST (Dermal Sensitization Threshold) model
TIMES DST model discriminates HPC (High Potency Category) from non-HPC chemicals with respect to dermal sensitization threshold accounting for (a)biotic activation of chemicals. It was developed by the Laboratory of Mathematical Chemistry using fundingfrom RIFM and with participation of Dr. David Roberts as reviewing expert.
The model was developed using a dataset of 557 unique chemicals tested by Local Lymph Node Assay (LLNA) and Human Repeat Insult Patch Test (HRIPT). All available skin sensitization data (EC3 and NOEL values in μg/cm2) from both tests were stored in the DST training set. In case of multiple data the worst case scenario (or expert judgement) is applied. Most potent data (EC3, NOEL) is selected to represent the chemical in the training set.As a result the DST database includes 557 unique chemicals distributed in three classess:- HPC - Non HPC (but positive sensitizers) - Weak/Non sensitizers
The TIMES DST model includes a list of protein binding alerts specified as Higly potent or Potent. The alerts are applied in combination with the TIMES simulators for abiotic oxidation (pre-electrophilic activation) and skin metabolism (pro-electrophilic activation). Because of thepaucity of reported skin metabolism data, initially the simulator transformations were developed based on empirical and theoretical knowledge. The transformation probabilities (defining the priority of their execution) were parameterized to reproduce skin sensitization data. Currently, the simulator was upgraded and adjusted to simulate the documented in vitro metabolism of 151 chemicals. The simulator comprises about 440 transformations, which can be divided into fourmain types:
- abiotic transformations
- covalent interaction with proteins
- Phase I and
- Phase II reactions.
Interactions with skin proteins are presented by 183 transformations grouped into two types: highly potent and potent. Depending of the applied protein binding alert TIMES DST model predicts compounds as HPC (High Potency Category) chemicals and non-HPC chemicals. Compounds for which no protein binding alert is found are predicted by the model as Weak/Non sensitizers.
The stepwise approach  was used to define the applicability domain of the model. It consists of the following sub-domain levels:
- General parametric requirements - includes ranges of variation log KOW and MW,
- Structural domain - based on atom-centered fragments (ACFs),
A chemical is considered In Domain if it is classified to belong to all sub-domain levels. The information implemented in the applicability domain is extracted from the correctly predicted training chemicals used to build the model and in this respect, the applicability domain determines practically the interpolation space of the model.
TIMES DST model was able to predict correctly 94% of HPC chemicals and 79% of non HPC chemicals, i.e., an overall performance of 81%.
The predictions from TIMES DST model could be reported as a tab delimited text file providingthe following information for the parent chemicals and their metabolites: chemical identity (CAS number, Name, SMILES), skin sensitization potency, applicability domain, reliability of protein binding alerts, etc.
S. Dimitrov, G. Dimitrova, T. Pavlov, N. Dimitrova, G. Patlevisz, J. Niemela and O.
Mekemyan, J. Chem. Inf. Model. 45 (2005), pp. 839-849.